xref: /openbmc/linux/net/sched/cls_api.c (revision 0545810f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * net/sched/cls_api.c	Packet classifier API.
4  *
5  * Authors:	Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6  *
7  * Changes:
8  *
9  * Eduardo J. Blanco <ejbs@netlabs.com.uy> :990222: kmod support
10  */
11 
12 #include <linux/module.h>
13 #include <linux/types.h>
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/err.h>
18 #include <linux/skbuff.h>
19 #include <linux/init.h>
20 #include <linux/kmod.h>
21 #include <linux/slab.h>
22 #include <linux/idr.h>
23 #include <linux/jhash.h>
24 #include <linux/rculist.h>
25 #include <linux/rhashtable.h>
26 #include <net/net_namespace.h>
27 #include <net/sock.h>
28 #include <net/netlink.h>
29 #include <net/pkt_sched.h>
30 #include <net/pkt_cls.h>
31 #include <net/tc_act/tc_pedit.h>
32 #include <net/tc_act/tc_mirred.h>
33 #include <net/tc_act/tc_vlan.h>
34 #include <net/tc_act/tc_tunnel_key.h>
35 #include <net/tc_act/tc_csum.h>
36 #include <net/tc_act/tc_gact.h>
37 #include <net/tc_act/tc_police.h>
38 #include <net/tc_act/tc_sample.h>
39 #include <net/tc_act/tc_skbedit.h>
40 #include <net/tc_act/tc_ct.h>
41 #include <net/tc_act/tc_mpls.h>
42 #include <net/tc_act/tc_gate.h>
43 #include <net/flow_offload.h>
44 #include <net/tc_wrapper.h>
45 
46 extern const struct nla_policy rtm_tca_policy[TCA_MAX + 1];
47 
48 /* The list of all installed classifier types */
49 static LIST_HEAD(tcf_proto_base);
50 
51 /* Protects list of registered TC modules. It is pure SMP lock. */
52 static DEFINE_RWLOCK(cls_mod_lock);
53 
54 static struct xarray tcf_exts_miss_cookies_xa;
55 struct tcf_exts_miss_cookie_node {
56 	const struct tcf_chain *chain;
57 	const struct tcf_proto *tp;
58 	const struct tcf_exts *exts;
59 	u32 chain_index;
60 	u32 tp_prio;
61 	u32 handle;
62 	u32 miss_cookie_base;
63 	struct rcu_head rcu;
64 };
65 
66 /* Each tc action entry cookie will be comprised of 32bit miss_cookie_base +
67  * action index in the exts tc actions array.
68  */
69 union tcf_exts_miss_cookie {
70 	struct {
71 		u32 miss_cookie_base;
72 		u32 act_index;
73 	};
74 	u64 miss_cookie;
75 };
76 
77 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
78 static int
79 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
80 				u32 handle)
81 {
82 	struct tcf_exts_miss_cookie_node *n;
83 	static u32 next;
84 	int err;
85 
86 	if (WARN_ON(!handle || !tp->ops->get_exts))
87 		return -EINVAL;
88 
89 	n = kzalloc(sizeof(*n), GFP_KERNEL);
90 	if (!n)
91 		return -ENOMEM;
92 
93 	n->chain_index = tp->chain->index;
94 	n->chain = tp->chain;
95 	n->tp_prio = tp->prio;
96 	n->tp = tp;
97 	n->exts = exts;
98 	n->handle = handle;
99 
100 	err = xa_alloc_cyclic(&tcf_exts_miss_cookies_xa, &n->miss_cookie_base,
101 			      n, xa_limit_32b, &next, GFP_KERNEL);
102 	if (err)
103 		goto err_xa_alloc;
104 
105 	exts->miss_cookie_node = n;
106 	return 0;
107 
108 err_xa_alloc:
109 	kfree(n);
110 	return err;
111 }
112 
113 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
114 {
115 	struct tcf_exts_miss_cookie_node *n;
116 
117 	if (!exts->miss_cookie_node)
118 		return;
119 
120 	n = exts->miss_cookie_node;
121 	xa_erase(&tcf_exts_miss_cookies_xa, n->miss_cookie_base);
122 	kfree_rcu(n, rcu);
123 }
124 
125 static struct tcf_exts_miss_cookie_node *
126 tcf_exts_miss_cookie_lookup(u64 miss_cookie, int *act_index)
127 {
128 	union tcf_exts_miss_cookie mc = { .miss_cookie = miss_cookie, };
129 
130 	*act_index = mc.act_index;
131 	return xa_load(&tcf_exts_miss_cookies_xa, mc.miss_cookie_base);
132 }
133 #else /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
134 static int
135 tcf_exts_miss_cookie_base_alloc(struct tcf_exts *exts, struct tcf_proto *tp,
136 				u32 handle)
137 {
138 	return 0;
139 }
140 
141 static void tcf_exts_miss_cookie_base_destroy(struct tcf_exts *exts)
142 {
143 }
144 #endif /* IS_ENABLED(CONFIG_NET_TC_SKB_EXT) */
145 
146 static u64 tcf_exts_miss_cookie_get(u32 miss_cookie_base, int act_index)
147 {
148 	union tcf_exts_miss_cookie mc = { .act_index = act_index, };
149 
150 	if (!miss_cookie_base)
151 		return 0;
152 
153 	mc.miss_cookie_base = miss_cookie_base;
154 	return mc.miss_cookie;
155 }
156 
157 #ifdef CONFIG_NET_CLS_ACT
158 DEFINE_STATIC_KEY_FALSE(tc_skb_ext_tc);
159 EXPORT_SYMBOL(tc_skb_ext_tc);
160 
161 void tc_skb_ext_tc_enable(void)
162 {
163 	static_branch_inc(&tc_skb_ext_tc);
164 }
165 EXPORT_SYMBOL(tc_skb_ext_tc_enable);
166 
167 void tc_skb_ext_tc_disable(void)
168 {
169 	static_branch_dec(&tc_skb_ext_tc);
170 }
171 EXPORT_SYMBOL(tc_skb_ext_tc_disable);
172 #endif
173 
174 static u32 destroy_obj_hashfn(const struct tcf_proto *tp)
175 {
176 	return jhash_3words(tp->chain->index, tp->prio,
177 			    (__force __u32)tp->protocol, 0);
178 }
179 
180 static void tcf_proto_signal_destroying(struct tcf_chain *chain,
181 					struct tcf_proto *tp)
182 {
183 	struct tcf_block *block = chain->block;
184 
185 	mutex_lock(&block->proto_destroy_lock);
186 	hash_add_rcu(block->proto_destroy_ht, &tp->destroy_ht_node,
187 		     destroy_obj_hashfn(tp));
188 	mutex_unlock(&block->proto_destroy_lock);
189 }
190 
191 static bool tcf_proto_cmp(const struct tcf_proto *tp1,
192 			  const struct tcf_proto *tp2)
193 {
194 	return tp1->chain->index == tp2->chain->index &&
195 	       tp1->prio == tp2->prio &&
196 	       tp1->protocol == tp2->protocol;
197 }
198 
199 static bool tcf_proto_exists_destroying(struct tcf_chain *chain,
200 					struct tcf_proto *tp)
201 {
202 	u32 hash = destroy_obj_hashfn(tp);
203 	struct tcf_proto *iter;
204 	bool found = false;
205 
206 	rcu_read_lock();
207 	hash_for_each_possible_rcu(chain->block->proto_destroy_ht, iter,
208 				   destroy_ht_node, hash) {
209 		if (tcf_proto_cmp(tp, iter)) {
210 			found = true;
211 			break;
212 		}
213 	}
214 	rcu_read_unlock();
215 
216 	return found;
217 }
218 
219 static void
220 tcf_proto_signal_destroyed(struct tcf_chain *chain, struct tcf_proto *tp)
221 {
222 	struct tcf_block *block = chain->block;
223 
224 	mutex_lock(&block->proto_destroy_lock);
225 	if (hash_hashed(&tp->destroy_ht_node))
226 		hash_del_rcu(&tp->destroy_ht_node);
227 	mutex_unlock(&block->proto_destroy_lock);
228 }
229 
230 /* Find classifier type by string name */
231 
232 static const struct tcf_proto_ops *__tcf_proto_lookup_ops(const char *kind)
233 {
234 	const struct tcf_proto_ops *t, *res = NULL;
235 
236 	if (kind) {
237 		read_lock(&cls_mod_lock);
238 		list_for_each_entry(t, &tcf_proto_base, head) {
239 			if (strcmp(kind, t->kind) == 0) {
240 				if (try_module_get(t->owner))
241 					res = t;
242 				break;
243 			}
244 		}
245 		read_unlock(&cls_mod_lock);
246 	}
247 	return res;
248 }
249 
250 static const struct tcf_proto_ops *
251 tcf_proto_lookup_ops(const char *kind, bool rtnl_held,
252 		     struct netlink_ext_ack *extack)
253 {
254 	const struct tcf_proto_ops *ops;
255 
256 	ops = __tcf_proto_lookup_ops(kind);
257 	if (ops)
258 		return ops;
259 #ifdef CONFIG_MODULES
260 	if (rtnl_held)
261 		rtnl_unlock();
262 	request_module("cls_%s", kind);
263 	if (rtnl_held)
264 		rtnl_lock();
265 	ops = __tcf_proto_lookup_ops(kind);
266 	/* We dropped the RTNL semaphore in order to perform
267 	 * the module load. So, even if we succeeded in loading
268 	 * the module we have to replay the request. We indicate
269 	 * this using -EAGAIN.
270 	 */
271 	if (ops) {
272 		module_put(ops->owner);
273 		return ERR_PTR(-EAGAIN);
274 	}
275 #endif
276 	NL_SET_ERR_MSG(extack, "TC classifier not found");
277 	return ERR_PTR(-ENOENT);
278 }
279 
280 /* Register(unregister) new classifier type */
281 
282 int register_tcf_proto_ops(struct tcf_proto_ops *ops)
283 {
284 	struct tcf_proto_ops *t;
285 	int rc = -EEXIST;
286 
287 	write_lock(&cls_mod_lock);
288 	list_for_each_entry(t, &tcf_proto_base, head)
289 		if (!strcmp(ops->kind, t->kind))
290 			goto out;
291 
292 	list_add_tail(&ops->head, &tcf_proto_base);
293 	rc = 0;
294 out:
295 	write_unlock(&cls_mod_lock);
296 	return rc;
297 }
298 EXPORT_SYMBOL(register_tcf_proto_ops);
299 
300 static struct workqueue_struct *tc_filter_wq;
301 
302 void unregister_tcf_proto_ops(struct tcf_proto_ops *ops)
303 {
304 	struct tcf_proto_ops *t;
305 	int rc = -ENOENT;
306 
307 	/* Wait for outstanding call_rcu()s, if any, from a
308 	 * tcf_proto_ops's destroy() handler.
309 	 */
310 	rcu_barrier();
311 	flush_workqueue(tc_filter_wq);
312 
313 	write_lock(&cls_mod_lock);
314 	list_for_each_entry(t, &tcf_proto_base, head) {
315 		if (t == ops) {
316 			list_del(&t->head);
317 			rc = 0;
318 			break;
319 		}
320 	}
321 	write_unlock(&cls_mod_lock);
322 
323 	WARN(rc, "unregister tc filter kind(%s) failed %d\n", ops->kind, rc);
324 }
325 EXPORT_SYMBOL(unregister_tcf_proto_ops);
326 
327 bool tcf_queue_work(struct rcu_work *rwork, work_func_t func)
328 {
329 	INIT_RCU_WORK(rwork, func);
330 	return queue_rcu_work(tc_filter_wq, rwork);
331 }
332 EXPORT_SYMBOL(tcf_queue_work);
333 
334 /* Select new prio value from the range, managed by kernel. */
335 
336 static inline u32 tcf_auto_prio(struct tcf_proto *tp)
337 {
338 	u32 first = TC_H_MAKE(0xC0000000U, 0U);
339 
340 	if (tp)
341 		first = tp->prio - 1;
342 
343 	return TC_H_MAJ(first);
344 }
345 
346 static bool tcf_proto_check_kind(struct nlattr *kind, char *name)
347 {
348 	if (kind)
349 		return nla_strscpy(name, kind, IFNAMSIZ) < 0;
350 	memset(name, 0, IFNAMSIZ);
351 	return false;
352 }
353 
354 static bool tcf_proto_is_unlocked(const char *kind)
355 {
356 	const struct tcf_proto_ops *ops;
357 	bool ret;
358 
359 	if (strlen(kind) == 0)
360 		return false;
361 
362 	ops = tcf_proto_lookup_ops(kind, false, NULL);
363 	/* On error return false to take rtnl lock. Proto lookup/create
364 	 * functions will perform lookup again and properly handle errors.
365 	 */
366 	if (IS_ERR(ops))
367 		return false;
368 
369 	ret = !!(ops->flags & TCF_PROTO_OPS_DOIT_UNLOCKED);
370 	module_put(ops->owner);
371 	return ret;
372 }
373 
374 static struct tcf_proto *tcf_proto_create(const char *kind, u32 protocol,
375 					  u32 prio, struct tcf_chain *chain,
376 					  bool rtnl_held,
377 					  struct netlink_ext_ack *extack)
378 {
379 	struct tcf_proto *tp;
380 	int err;
381 
382 	tp = kzalloc(sizeof(*tp), GFP_KERNEL);
383 	if (!tp)
384 		return ERR_PTR(-ENOBUFS);
385 
386 	tp->ops = tcf_proto_lookup_ops(kind, rtnl_held, extack);
387 	if (IS_ERR(tp->ops)) {
388 		err = PTR_ERR(tp->ops);
389 		goto errout;
390 	}
391 	tp->classify = tp->ops->classify;
392 	tp->protocol = protocol;
393 	tp->prio = prio;
394 	tp->chain = chain;
395 	spin_lock_init(&tp->lock);
396 	refcount_set(&tp->refcnt, 1);
397 
398 	err = tp->ops->init(tp);
399 	if (err) {
400 		module_put(tp->ops->owner);
401 		goto errout;
402 	}
403 	return tp;
404 
405 errout:
406 	kfree(tp);
407 	return ERR_PTR(err);
408 }
409 
410 static void tcf_proto_get(struct tcf_proto *tp)
411 {
412 	refcount_inc(&tp->refcnt);
413 }
414 
415 static void tcf_chain_put(struct tcf_chain *chain);
416 
417 static void tcf_proto_destroy(struct tcf_proto *tp, bool rtnl_held,
418 			      bool sig_destroy, struct netlink_ext_ack *extack)
419 {
420 	tp->ops->destroy(tp, rtnl_held, extack);
421 	if (sig_destroy)
422 		tcf_proto_signal_destroyed(tp->chain, tp);
423 	tcf_chain_put(tp->chain);
424 	module_put(tp->ops->owner);
425 	kfree_rcu(tp, rcu);
426 }
427 
428 static void tcf_proto_put(struct tcf_proto *tp, bool rtnl_held,
429 			  struct netlink_ext_ack *extack)
430 {
431 	if (refcount_dec_and_test(&tp->refcnt))
432 		tcf_proto_destroy(tp, rtnl_held, true, extack);
433 }
434 
435 static bool tcf_proto_check_delete(struct tcf_proto *tp)
436 {
437 	if (tp->ops->delete_empty)
438 		return tp->ops->delete_empty(tp);
439 
440 	tp->deleting = true;
441 	return tp->deleting;
442 }
443 
444 static void tcf_proto_mark_delete(struct tcf_proto *tp)
445 {
446 	spin_lock(&tp->lock);
447 	tp->deleting = true;
448 	spin_unlock(&tp->lock);
449 }
450 
451 static bool tcf_proto_is_deleting(struct tcf_proto *tp)
452 {
453 	bool deleting;
454 
455 	spin_lock(&tp->lock);
456 	deleting = tp->deleting;
457 	spin_unlock(&tp->lock);
458 
459 	return deleting;
460 }
461 
462 #define ASSERT_BLOCK_LOCKED(block)					\
463 	lockdep_assert_held(&(block)->lock)
464 
465 struct tcf_filter_chain_list_item {
466 	struct list_head list;
467 	tcf_chain_head_change_t *chain_head_change;
468 	void *chain_head_change_priv;
469 };
470 
471 static struct tcf_chain *tcf_chain_create(struct tcf_block *block,
472 					  u32 chain_index)
473 {
474 	struct tcf_chain *chain;
475 
476 	ASSERT_BLOCK_LOCKED(block);
477 
478 	chain = kzalloc(sizeof(*chain), GFP_KERNEL);
479 	if (!chain)
480 		return NULL;
481 	list_add_tail_rcu(&chain->list, &block->chain_list);
482 	mutex_init(&chain->filter_chain_lock);
483 	chain->block = block;
484 	chain->index = chain_index;
485 	chain->refcnt = 1;
486 	if (!chain->index)
487 		block->chain0.chain = chain;
488 	return chain;
489 }
490 
491 static void tcf_chain_head_change_item(struct tcf_filter_chain_list_item *item,
492 				       struct tcf_proto *tp_head)
493 {
494 	if (item->chain_head_change)
495 		item->chain_head_change(tp_head, item->chain_head_change_priv);
496 }
497 
498 static void tcf_chain0_head_change(struct tcf_chain *chain,
499 				   struct tcf_proto *tp_head)
500 {
501 	struct tcf_filter_chain_list_item *item;
502 	struct tcf_block *block = chain->block;
503 
504 	if (chain->index)
505 		return;
506 
507 	mutex_lock(&block->lock);
508 	list_for_each_entry(item, &block->chain0.filter_chain_list, list)
509 		tcf_chain_head_change_item(item, tp_head);
510 	mutex_unlock(&block->lock);
511 }
512 
513 /* Returns true if block can be safely freed. */
514 
515 static bool tcf_chain_detach(struct tcf_chain *chain)
516 {
517 	struct tcf_block *block = chain->block;
518 
519 	ASSERT_BLOCK_LOCKED(block);
520 
521 	list_del_rcu(&chain->list);
522 	if (!chain->index)
523 		block->chain0.chain = NULL;
524 
525 	if (list_empty(&block->chain_list) &&
526 	    refcount_read(&block->refcnt) == 0)
527 		return true;
528 
529 	return false;
530 }
531 
532 static void tcf_block_destroy(struct tcf_block *block)
533 {
534 	mutex_destroy(&block->lock);
535 	mutex_destroy(&block->proto_destroy_lock);
536 	kfree_rcu(block, rcu);
537 }
538 
539 static void tcf_chain_destroy(struct tcf_chain *chain, bool free_block)
540 {
541 	struct tcf_block *block = chain->block;
542 
543 	mutex_destroy(&chain->filter_chain_lock);
544 	kfree_rcu(chain, rcu);
545 	if (free_block)
546 		tcf_block_destroy(block);
547 }
548 
549 static void tcf_chain_hold(struct tcf_chain *chain)
550 {
551 	ASSERT_BLOCK_LOCKED(chain->block);
552 
553 	++chain->refcnt;
554 }
555 
556 static bool tcf_chain_held_by_acts_only(struct tcf_chain *chain)
557 {
558 	ASSERT_BLOCK_LOCKED(chain->block);
559 
560 	/* In case all the references are action references, this
561 	 * chain should not be shown to the user.
562 	 */
563 	return chain->refcnt == chain->action_refcnt;
564 }
565 
566 static struct tcf_chain *tcf_chain_lookup(struct tcf_block *block,
567 					  u32 chain_index)
568 {
569 	struct tcf_chain *chain;
570 
571 	ASSERT_BLOCK_LOCKED(block);
572 
573 	list_for_each_entry(chain, &block->chain_list, list) {
574 		if (chain->index == chain_index)
575 			return chain;
576 	}
577 	return NULL;
578 }
579 
580 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
581 static struct tcf_chain *tcf_chain_lookup_rcu(const struct tcf_block *block,
582 					      u32 chain_index)
583 {
584 	struct tcf_chain *chain;
585 
586 	list_for_each_entry_rcu(chain, &block->chain_list, list) {
587 		if (chain->index == chain_index)
588 			return chain;
589 	}
590 	return NULL;
591 }
592 #endif
593 
594 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
595 			   u32 seq, u16 flags, int event, bool unicast,
596 			   struct netlink_ext_ack *extack);
597 
598 static struct tcf_chain *__tcf_chain_get(struct tcf_block *block,
599 					 u32 chain_index, bool create,
600 					 bool by_act)
601 {
602 	struct tcf_chain *chain = NULL;
603 	bool is_first_reference;
604 
605 	mutex_lock(&block->lock);
606 	chain = tcf_chain_lookup(block, chain_index);
607 	if (chain) {
608 		tcf_chain_hold(chain);
609 	} else {
610 		if (!create)
611 			goto errout;
612 		chain = tcf_chain_create(block, chain_index);
613 		if (!chain)
614 			goto errout;
615 	}
616 
617 	if (by_act)
618 		++chain->action_refcnt;
619 	is_first_reference = chain->refcnt - chain->action_refcnt == 1;
620 	mutex_unlock(&block->lock);
621 
622 	/* Send notification only in case we got the first
623 	 * non-action reference. Until then, the chain acts only as
624 	 * a placeholder for actions pointing to it and user ought
625 	 * not know about them.
626 	 */
627 	if (is_first_reference && !by_act)
628 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
629 				RTM_NEWCHAIN, false, NULL);
630 
631 	return chain;
632 
633 errout:
634 	mutex_unlock(&block->lock);
635 	return chain;
636 }
637 
638 static struct tcf_chain *tcf_chain_get(struct tcf_block *block, u32 chain_index,
639 				       bool create)
640 {
641 	return __tcf_chain_get(block, chain_index, create, false);
642 }
643 
644 struct tcf_chain *tcf_chain_get_by_act(struct tcf_block *block, u32 chain_index)
645 {
646 	return __tcf_chain_get(block, chain_index, true, true);
647 }
648 EXPORT_SYMBOL(tcf_chain_get_by_act);
649 
650 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
651 			       void *tmplt_priv);
652 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
653 				  void *tmplt_priv, u32 chain_index,
654 				  struct tcf_block *block, struct sk_buff *oskb,
655 				  u32 seq, u16 flags, bool unicast);
656 
657 static void __tcf_chain_put(struct tcf_chain *chain, bool by_act,
658 			    bool explicitly_created)
659 {
660 	struct tcf_block *block = chain->block;
661 	const struct tcf_proto_ops *tmplt_ops;
662 	bool free_block = false;
663 	unsigned int refcnt;
664 	void *tmplt_priv;
665 
666 	mutex_lock(&block->lock);
667 	if (explicitly_created) {
668 		if (!chain->explicitly_created) {
669 			mutex_unlock(&block->lock);
670 			return;
671 		}
672 		chain->explicitly_created = false;
673 	}
674 
675 	if (by_act)
676 		chain->action_refcnt--;
677 
678 	/* tc_chain_notify_delete can't be called while holding block lock.
679 	 * However, when block is unlocked chain can be changed concurrently, so
680 	 * save these to temporary variables.
681 	 */
682 	refcnt = --chain->refcnt;
683 	tmplt_ops = chain->tmplt_ops;
684 	tmplt_priv = chain->tmplt_priv;
685 
686 	/* The last dropped non-action reference will trigger notification. */
687 	if (refcnt - chain->action_refcnt == 0 && !by_act) {
688 		tc_chain_notify_delete(tmplt_ops, tmplt_priv, chain->index,
689 				       block, NULL, 0, 0, false);
690 		/* Last reference to chain, no need to lock. */
691 		chain->flushing = false;
692 	}
693 
694 	if (refcnt == 0)
695 		free_block = tcf_chain_detach(chain);
696 	mutex_unlock(&block->lock);
697 
698 	if (refcnt == 0) {
699 		tc_chain_tmplt_del(tmplt_ops, tmplt_priv);
700 		tcf_chain_destroy(chain, free_block);
701 	}
702 }
703 
704 static void tcf_chain_put(struct tcf_chain *chain)
705 {
706 	__tcf_chain_put(chain, false, false);
707 }
708 
709 void tcf_chain_put_by_act(struct tcf_chain *chain)
710 {
711 	__tcf_chain_put(chain, true, false);
712 }
713 EXPORT_SYMBOL(tcf_chain_put_by_act);
714 
715 static void tcf_chain_put_explicitly_created(struct tcf_chain *chain)
716 {
717 	__tcf_chain_put(chain, false, true);
718 }
719 
720 static void tcf_chain_flush(struct tcf_chain *chain, bool rtnl_held)
721 {
722 	struct tcf_proto *tp, *tp_next;
723 
724 	mutex_lock(&chain->filter_chain_lock);
725 	tp = tcf_chain_dereference(chain->filter_chain, chain);
726 	while (tp) {
727 		tp_next = rcu_dereference_protected(tp->next, 1);
728 		tcf_proto_signal_destroying(chain, tp);
729 		tp = tp_next;
730 	}
731 	tp = tcf_chain_dereference(chain->filter_chain, chain);
732 	RCU_INIT_POINTER(chain->filter_chain, NULL);
733 	tcf_chain0_head_change(chain, NULL);
734 	chain->flushing = true;
735 	mutex_unlock(&chain->filter_chain_lock);
736 
737 	while (tp) {
738 		tp_next = rcu_dereference_protected(tp->next, 1);
739 		tcf_proto_put(tp, rtnl_held, NULL);
740 		tp = tp_next;
741 	}
742 }
743 
744 static int tcf_block_setup(struct tcf_block *block,
745 			   struct flow_block_offload *bo);
746 
747 static void tcf_block_offload_init(struct flow_block_offload *bo,
748 				   struct net_device *dev, struct Qdisc *sch,
749 				   enum flow_block_command command,
750 				   enum flow_block_binder_type binder_type,
751 				   struct flow_block *flow_block,
752 				   bool shared, struct netlink_ext_ack *extack)
753 {
754 	bo->net = dev_net(dev);
755 	bo->command = command;
756 	bo->binder_type = binder_type;
757 	bo->block = flow_block;
758 	bo->block_shared = shared;
759 	bo->extack = extack;
760 	bo->sch = sch;
761 	bo->cb_list_head = &flow_block->cb_list;
762 	INIT_LIST_HEAD(&bo->cb_list);
763 }
764 
765 static void tcf_block_unbind(struct tcf_block *block,
766 			     struct flow_block_offload *bo);
767 
768 static void tc_block_indr_cleanup(struct flow_block_cb *block_cb)
769 {
770 	struct tcf_block *block = block_cb->indr.data;
771 	struct net_device *dev = block_cb->indr.dev;
772 	struct Qdisc *sch = block_cb->indr.sch;
773 	struct netlink_ext_ack extack = {};
774 	struct flow_block_offload bo = {};
775 
776 	tcf_block_offload_init(&bo, dev, sch, FLOW_BLOCK_UNBIND,
777 			       block_cb->indr.binder_type,
778 			       &block->flow_block, tcf_block_shared(block),
779 			       &extack);
780 	rtnl_lock();
781 	down_write(&block->cb_lock);
782 	list_del(&block_cb->driver_list);
783 	list_move(&block_cb->list, &bo.cb_list);
784 	tcf_block_unbind(block, &bo);
785 	up_write(&block->cb_lock);
786 	rtnl_unlock();
787 }
788 
789 static bool tcf_block_offload_in_use(struct tcf_block *block)
790 {
791 	return atomic_read(&block->offloadcnt);
792 }
793 
794 static int tcf_block_offload_cmd(struct tcf_block *block,
795 				 struct net_device *dev, struct Qdisc *sch,
796 				 struct tcf_block_ext_info *ei,
797 				 enum flow_block_command command,
798 				 struct netlink_ext_ack *extack)
799 {
800 	struct flow_block_offload bo = {};
801 
802 	tcf_block_offload_init(&bo, dev, sch, command, ei->binder_type,
803 			       &block->flow_block, tcf_block_shared(block),
804 			       extack);
805 
806 	if (dev->netdev_ops->ndo_setup_tc) {
807 		int err;
808 
809 		err = dev->netdev_ops->ndo_setup_tc(dev, TC_SETUP_BLOCK, &bo);
810 		if (err < 0) {
811 			if (err != -EOPNOTSUPP)
812 				NL_SET_ERR_MSG(extack, "Driver ndo_setup_tc failed");
813 			return err;
814 		}
815 
816 		return tcf_block_setup(block, &bo);
817 	}
818 
819 	flow_indr_dev_setup_offload(dev, sch, TC_SETUP_BLOCK, block, &bo,
820 				    tc_block_indr_cleanup);
821 	tcf_block_setup(block, &bo);
822 
823 	return -EOPNOTSUPP;
824 }
825 
826 static int tcf_block_offload_bind(struct tcf_block *block, struct Qdisc *q,
827 				  struct tcf_block_ext_info *ei,
828 				  struct netlink_ext_ack *extack)
829 {
830 	struct net_device *dev = q->dev_queue->dev;
831 	int err;
832 
833 	down_write(&block->cb_lock);
834 
835 	/* If tc offload feature is disabled and the block we try to bind
836 	 * to already has some offloaded filters, forbid to bind.
837 	 */
838 	if (dev->netdev_ops->ndo_setup_tc &&
839 	    !tc_can_offload(dev) &&
840 	    tcf_block_offload_in_use(block)) {
841 		NL_SET_ERR_MSG(extack, "Bind to offloaded block failed as dev has offload disabled");
842 		err = -EOPNOTSUPP;
843 		goto err_unlock;
844 	}
845 
846 	err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_BIND, extack);
847 	if (err == -EOPNOTSUPP)
848 		goto no_offload_dev_inc;
849 	if (err)
850 		goto err_unlock;
851 
852 	up_write(&block->cb_lock);
853 	return 0;
854 
855 no_offload_dev_inc:
856 	if (tcf_block_offload_in_use(block))
857 		goto err_unlock;
858 
859 	err = 0;
860 	block->nooffloaddevcnt++;
861 err_unlock:
862 	up_write(&block->cb_lock);
863 	return err;
864 }
865 
866 static void tcf_block_offload_unbind(struct tcf_block *block, struct Qdisc *q,
867 				     struct tcf_block_ext_info *ei)
868 {
869 	struct net_device *dev = q->dev_queue->dev;
870 	int err;
871 
872 	down_write(&block->cb_lock);
873 	err = tcf_block_offload_cmd(block, dev, q, ei, FLOW_BLOCK_UNBIND, NULL);
874 	if (err == -EOPNOTSUPP)
875 		goto no_offload_dev_dec;
876 	up_write(&block->cb_lock);
877 	return;
878 
879 no_offload_dev_dec:
880 	WARN_ON(block->nooffloaddevcnt-- == 0);
881 	up_write(&block->cb_lock);
882 }
883 
884 static int
885 tcf_chain0_head_change_cb_add(struct tcf_block *block,
886 			      struct tcf_block_ext_info *ei,
887 			      struct netlink_ext_ack *extack)
888 {
889 	struct tcf_filter_chain_list_item *item;
890 	struct tcf_chain *chain0;
891 
892 	item = kmalloc(sizeof(*item), GFP_KERNEL);
893 	if (!item) {
894 		NL_SET_ERR_MSG(extack, "Memory allocation for head change callback item failed");
895 		return -ENOMEM;
896 	}
897 	item->chain_head_change = ei->chain_head_change;
898 	item->chain_head_change_priv = ei->chain_head_change_priv;
899 
900 	mutex_lock(&block->lock);
901 	chain0 = block->chain0.chain;
902 	if (chain0)
903 		tcf_chain_hold(chain0);
904 	else
905 		list_add(&item->list, &block->chain0.filter_chain_list);
906 	mutex_unlock(&block->lock);
907 
908 	if (chain0) {
909 		struct tcf_proto *tp_head;
910 
911 		mutex_lock(&chain0->filter_chain_lock);
912 
913 		tp_head = tcf_chain_dereference(chain0->filter_chain, chain0);
914 		if (tp_head)
915 			tcf_chain_head_change_item(item, tp_head);
916 
917 		mutex_lock(&block->lock);
918 		list_add(&item->list, &block->chain0.filter_chain_list);
919 		mutex_unlock(&block->lock);
920 
921 		mutex_unlock(&chain0->filter_chain_lock);
922 		tcf_chain_put(chain0);
923 	}
924 
925 	return 0;
926 }
927 
928 static void
929 tcf_chain0_head_change_cb_del(struct tcf_block *block,
930 			      struct tcf_block_ext_info *ei)
931 {
932 	struct tcf_filter_chain_list_item *item;
933 
934 	mutex_lock(&block->lock);
935 	list_for_each_entry(item, &block->chain0.filter_chain_list, list) {
936 		if ((!ei->chain_head_change && !ei->chain_head_change_priv) ||
937 		    (item->chain_head_change == ei->chain_head_change &&
938 		     item->chain_head_change_priv == ei->chain_head_change_priv)) {
939 			if (block->chain0.chain)
940 				tcf_chain_head_change_item(item, NULL);
941 			list_del(&item->list);
942 			mutex_unlock(&block->lock);
943 
944 			kfree(item);
945 			return;
946 		}
947 	}
948 	mutex_unlock(&block->lock);
949 	WARN_ON(1);
950 }
951 
952 struct tcf_net {
953 	spinlock_t idr_lock; /* Protects idr */
954 	struct idr idr;
955 };
956 
957 static unsigned int tcf_net_id;
958 
959 static int tcf_block_insert(struct tcf_block *block, struct net *net,
960 			    struct netlink_ext_ack *extack)
961 {
962 	struct tcf_net *tn = net_generic(net, tcf_net_id);
963 	int err;
964 
965 	idr_preload(GFP_KERNEL);
966 	spin_lock(&tn->idr_lock);
967 	err = idr_alloc_u32(&tn->idr, block, &block->index, block->index,
968 			    GFP_NOWAIT);
969 	spin_unlock(&tn->idr_lock);
970 	idr_preload_end();
971 
972 	return err;
973 }
974 
975 static void tcf_block_remove(struct tcf_block *block, struct net *net)
976 {
977 	struct tcf_net *tn = net_generic(net, tcf_net_id);
978 
979 	spin_lock(&tn->idr_lock);
980 	idr_remove(&tn->idr, block->index);
981 	spin_unlock(&tn->idr_lock);
982 }
983 
984 static struct tcf_block *tcf_block_create(struct net *net, struct Qdisc *q,
985 					  u32 block_index,
986 					  struct netlink_ext_ack *extack)
987 {
988 	struct tcf_block *block;
989 
990 	block = kzalloc(sizeof(*block), GFP_KERNEL);
991 	if (!block) {
992 		NL_SET_ERR_MSG(extack, "Memory allocation for block failed");
993 		return ERR_PTR(-ENOMEM);
994 	}
995 	mutex_init(&block->lock);
996 	mutex_init(&block->proto_destroy_lock);
997 	init_rwsem(&block->cb_lock);
998 	flow_block_init(&block->flow_block);
999 	INIT_LIST_HEAD(&block->chain_list);
1000 	INIT_LIST_HEAD(&block->owner_list);
1001 	INIT_LIST_HEAD(&block->chain0.filter_chain_list);
1002 
1003 	refcount_set(&block->refcnt, 1);
1004 	block->net = net;
1005 	block->index = block_index;
1006 
1007 	/* Don't store q pointer for blocks which are shared */
1008 	if (!tcf_block_shared(block))
1009 		block->q = q;
1010 	return block;
1011 }
1012 
1013 static struct tcf_block *tcf_block_lookup(struct net *net, u32 block_index)
1014 {
1015 	struct tcf_net *tn = net_generic(net, tcf_net_id);
1016 
1017 	return idr_find(&tn->idr, block_index);
1018 }
1019 
1020 static struct tcf_block *tcf_block_refcnt_get(struct net *net, u32 block_index)
1021 {
1022 	struct tcf_block *block;
1023 
1024 	rcu_read_lock();
1025 	block = tcf_block_lookup(net, block_index);
1026 	if (block && !refcount_inc_not_zero(&block->refcnt))
1027 		block = NULL;
1028 	rcu_read_unlock();
1029 
1030 	return block;
1031 }
1032 
1033 static struct tcf_chain *
1034 __tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1035 {
1036 	mutex_lock(&block->lock);
1037 	if (chain)
1038 		chain = list_is_last(&chain->list, &block->chain_list) ?
1039 			NULL : list_next_entry(chain, list);
1040 	else
1041 		chain = list_first_entry_or_null(&block->chain_list,
1042 						 struct tcf_chain, list);
1043 
1044 	/* skip all action-only chains */
1045 	while (chain && tcf_chain_held_by_acts_only(chain))
1046 		chain = list_is_last(&chain->list, &block->chain_list) ?
1047 			NULL : list_next_entry(chain, list);
1048 
1049 	if (chain)
1050 		tcf_chain_hold(chain);
1051 	mutex_unlock(&block->lock);
1052 
1053 	return chain;
1054 }
1055 
1056 /* Function to be used by all clients that want to iterate over all chains on
1057  * block. It properly obtains block->lock and takes reference to chain before
1058  * returning it. Users of this function must be tolerant to concurrent chain
1059  * insertion/deletion or ensure that no concurrent chain modification is
1060  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1061  * consistent dump because rtnl lock is released each time skb is filled with
1062  * data and sent to user-space.
1063  */
1064 
1065 struct tcf_chain *
1066 tcf_get_next_chain(struct tcf_block *block, struct tcf_chain *chain)
1067 {
1068 	struct tcf_chain *chain_next = __tcf_get_next_chain(block, chain);
1069 
1070 	if (chain)
1071 		tcf_chain_put(chain);
1072 
1073 	return chain_next;
1074 }
1075 EXPORT_SYMBOL(tcf_get_next_chain);
1076 
1077 static struct tcf_proto *
1078 __tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1079 {
1080 	u32 prio = 0;
1081 
1082 	ASSERT_RTNL();
1083 	mutex_lock(&chain->filter_chain_lock);
1084 
1085 	if (!tp) {
1086 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1087 	} else if (tcf_proto_is_deleting(tp)) {
1088 		/* 'deleting' flag is set and chain->filter_chain_lock was
1089 		 * unlocked, which means next pointer could be invalid. Restart
1090 		 * search.
1091 		 */
1092 		prio = tp->prio + 1;
1093 		tp = tcf_chain_dereference(chain->filter_chain, chain);
1094 
1095 		for (; tp; tp = tcf_chain_dereference(tp->next, chain))
1096 			if (!tp->deleting && tp->prio >= prio)
1097 				break;
1098 	} else {
1099 		tp = tcf_chain_dereference(tp->next, chain);
1100 	}
1101 
1102 	if (tp)
1103 		tcf_proto_get(tp);
1104 
1105 	mutex_unlock(&chain->filter_chain_lock);
1106 
1107 	return tp;
1108 }
1109 
1110 /* Function to be used by all clients that want to iterate over all tp's on
1111  * chain. Users of this function must be tolerant to concurrent tp
1112  * insertion/deletion or ensure that no concurrent chain modification is
1113  * possible. Note that all netlink dump callbacks cannot guarantee to provide
1114  * consistent dump because rtnl lock is released each time skb is filled with
1115  * data and sent to user-space.
1116  */
1117 
1118 struct tcf_proto *
1119 tcf_get_next_proto(struct tcf_chain *chain, struct tcf_proto *tp)
1120 {
1121 	struct tcf_proto *tp_next = __tcf_get_next_proto(chain, tp);
1122 
1123 	if (tp)
1124 		tcf_proto_put(tp, true, NULL);
1125 
1126 	return tp_next;
1127 }
1128 EXPORT_SYMBOL(tcf_get_next_proto);
1129 
1130 static void tcf_block_flush_all_chains(struct tcf_block *block, bool rtnl_held)
1131 {
1132 	struct tcf_chain *chain;
1133 
1134 	/* Last reference to block. At this point chains cannot be added or
1135 	 * removed concurrently.
1136 	 */
1137 	for (chain = tcf_get_next_chain(block, NULL);
1138 	     chain;
1139 	     chain = tcf_get_next_chain(block, chain)) {
1140 		tcf_chain_put_explicitly_created(chain);
1141 		tcf_chain_flush(chain, rtnl_held);
1142 	}
1143 }
1144 
1145 /* Lookup Qdisc and increments its reference counter.
1146  * Set parent, if necessary.
1147  */
1148 
1149 static int __tcf_qdisc_find(struct net *net, struct Qdisc **q,
1150 			    u32 *parent, int ifindex, bool rtnl_held,
1151 			    struct netlink_ext_ack *extack)
1152 {
1153 	const struct Qdisc_class_ops *cops;
1154 	struct net_device *dev;
1155 	int err = 0;
1156 
1157 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1158 		return 0;
1159 
1160 	rcu_read_lock();
1161 
1162 	/* Find link */
1163 	dev = dev_get_by_index_rcu(net, ifindex);
1164 	if (!dev) {
1165 		rcu_read_unlock();
1166 		return -ENODEV;
1167 	}
1168 
1169 	/* Find qdisc */
1170 	if (!*parent) {
1171 		*q = rcu_dereference(dev->qdisc);
1172 		*parent = (*q)->handle;
1173 	} else {
1174 		*q = qdisc_lookup_rcu(dev, TC_H_MAJ(*parent));
1175 		if (!*q) {
1176 			NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1177 			err = -EINVAL;
1178 			goto errout_rcu;
1179 		}
1180 	}
1181 
1182 	*q = qdisc_refcount_inc_nz(*q);
1183 	if (!*q) {
1184 		NL_SET_ERR_MSG(extack, "Parent Qdisc doesn't exists");
1185 		err = -EINVAL;
1186 		goto errout_rcu;
1187 	}
1188 
1189 	/* Is it classful? */
1190 	cops = (*q)->ops->cl_ops;
1191 	if (!cops) {
1192 		NL_SET_ERR_MSG(extack, "Qdisc not classful");
1193 		err = -EINVAL;
1194 		goto errout_qdisc;
1195 	}
1196 
1197 	if (!cops->tcf_block) {
1198 		NL_SET_ERR_MSG(extack, "Class doesn't support blocks");
1199 		err = -EOPNOTSUPP;
1200 		goto errout_qdisc;
1201 	}
1202 
1203 errout_rcu:
1204 	/* At this point we know that qdisc is not noop_qdisc,
1205 	 * which means that qdisc holds a reference to net_device
1206 	 * and we hold a reference to qdisc, so it is safe to release
1207 	 * rcu read lock.
1208 	 */
1209 	rcu_read_unlock();
1210 	return err;
1211 
1212 errout_qdisc:
1213 	rcu_read_unlock();
1214 
1215 	if (rtnl_held)
1216 		qdisc_put(*q);
1217 	else
1218 		qdisc_put_unlocked(*q);
1219 	*q = NULL;
1220 
1221 	return err;
1222 }
1223 
1224 static int __tcf_qdisc_cl_find(struct Qdisc *q, u32 parent, unsigned long *cl,
1225 			       int ifindex, struct netlink_ext_ack *extack)
1226 {
1227 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK)
1228 		return 0;
1229 
1230 	/* Do we search for filter, attached to class? */
1231 	if (TC_H_MIN(parent)) {
1232 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1233 
1234 		*cl = cops->find(q, parent);
1235 		if (*cl == 0) {
1236 			NL_SET_ERR_MSG(extack, "Specified class doesn't exist");
1237 			return -ENOENT;
1238 		}
1239 	}
1240 
1241 	return 0;
1242 }
1243 
1244 static struct tcf_block *__tcf_block_find(struct net *net, struct Qdisc *q,
1245 					  unsigned long cl, int ifindex,
1246 					  u32 block_index,
1247 					  struct netlink_ext_ack *extack)
1248 {
1249 	struct tcf_block *block;
1250 
1251 	if (ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
1252 		block = tcf_block_refcnt_get(net, block_index);
1253 		if (!block) {
1254 			NL_SET_ERR_MSG(extack, "Block of given index was not found");
1255 			return ERR_PTR(-EINVAL);
1256 		}
1257 	} else {
1258 		const struct Qdisc_class_ops *cops = q->ops->cl_ops;
1259 
1260 		block = cops->tcf_block(q, cl, extack);
1261 		if (!block)
1262 			return ERR_PTR(-EINVAL);
1263 
1264 		if (tcf_block_shared(block)) {
1265 			NL_SET_ERR_MSG(extack, "This filter block is shared. Please use the block index to manipulate the filters");
1266 			return ERR_PTR(-EOPNOTSUPP);
1267 		}
1268 
1269 		/* Always take reference to block in order to support execution
1270 		 * of rules update path of cls API without rtnl lock. Caller
1271 		 * must release block when it is finished using it. 'if' block
1272 		 * of this conditional obtain reference to block by calling
1273 		 * tcf_block_refcnt_get().
1274 		 */
1275 		refcount_inc(&block->refcnt);
1276 	}
1277 
1278 	return block;
1279 }
1280 
1281 static void __tcf_block_put(struct tcf_block *block, struct Qdisc *q,
1282 			    struct tcf_block_ext_info *ei, bool rtnl_held)
1283 {
1284 	if (refcount_dec_and_mutex_lock(&block->refcnt, &block->lock)) {
1285 		/* Flushing/putting all chains will cause the block to be
1286 		 * deallocated when last chain is freed. However, if chain_list
1287 		 * is empty, block has to be manually deallocated. After block
1288 		 * reference counter reached 0, it is no longer possible to
1289 		 * increment it or add new chains to block.
1290 		 */
1291 		bool free_block = list_empty(&block->chain_list);
1292 
1293 		mutex_unlock(&block->lock);
1294 		if (tcf_block_shared(block))
1295 			tcf_block_remove(block, block->net);
1296 
1297 		if (q)
1298 			tcf_block_offload_unbind(block, q, ei);
1299 
1300 		if (free_block)
1301 			tcf_block_destroy(block);
1302 		else
1303 			tcf_block_flush_all_chains(block, rtnl_held);
1304 	} else if (q) {
1305 		tcf_block_offload_unbind(block, q, ei);
1306 	}
1307 }
1308 
1309 static void tcf_block_refcnt_put(struct tcf_block *block, bool rtnl_held)
1310 {
1311 	__tcf_block_put(block, NULL, NULL, rtnl_held);
1312 }
1313 
1314 /* Find tcf block.
1315  * Set q, parent, cl when appropriate.
1316  */
1317 
1318 static struct tcf_block *tcf_block_find(struct net *net, struct Qdisc **q,
1319 					u32 *parent, unsigned long *cl,
1320 					int ifindex, u32 block_index,
1321 					struct netlink_ext_ack *extack)
1322 {
1323 	struct tcf_block *block;
1324 	int err = 0;
1325 
1326 	ASSERT_RTNL();
1327 
1328 	err = __tcf_qdisc_find(net, q, parent, ifindex, true, extack);
1329 	if (err)
1330 		goto errout;
1331 
1332 	err = __tcf_qdisc_cl_find(*q, *parent, cl, ifindex, extack);
1333 	if (err)
1334 		goto errout_qdisc;
1335 
1336 	block = __tcf_block_find(net, *q, *cl, ifindex, block_index, extack);
1337 	if (IS_ERR(block)) {
1338 		err = PTR_ERR(block);
1339 		goto errout_qdisc;
1340 	}
1341 
1342 	return block;
1343 
1344 errout_qdisc:
1345 	if (*q)
1346 		qdisc_put(*q);
1347 errout:
1348 	*q = NULL;
1349 	return ERR_PTR(err);
1350 }
1351 
1352 static void tcf_block_release(struct Qdisc *q, struct tcf_block *block,
1353 			      bool rtnl_held)
1354 {
1355 	if (!IS_ERR_OR_NULL(block))
1356 		tcf_block_refcnt_put(block, rtnl_held);
1357 
1358 	if (q) {
1359 		if (rtnl_held)
1360 			qdisc_put(q);
1361 		else
1362 			qdisc_put_unlocked(q);
1363 	}
1364 }
1365 
1366 struct tcf_block_owner_item {
1367 	struct list_head list;
1368 	struct Qdisc *q;
1369 	enum flow_block_binder_type binder_type;
1370 };
1371 
1372 static void
1373 tcf_block_owner_netif_keep_dst(struct tcf_block *block,
1374 			       struct Qdisc *q,
1375 			       enum flow_block_binder_type binder_type)
1376 {
1377 	if (block->keep_dst &&
1378 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS &&
1379 	    binder_type != FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS)
1380 		netif_keep_dst(qdisc_dev(q));
1381 }
1382 
1383 void tcf_block_netif_keep_dst(struct tcf_block *block)
1384 {
1385 	struct tcf_block_owner_item *item;
1386 
1387 	block->keep_dst = true;
1388 	list_for_each_entry(item, &block->owner_list, list)
1389 		tcf_block_owner_netif_keep_dst(block, item->q,
1390 					       item->binder_type);
1391 }
1392 EXPORT_SYMBOL(tcf_block_netif_keep_dst);
1393 
1394 static int tcf_block_owner_add(struct tcf_block *block,
1395 			       struct Qdisc *q,
1396 			       enum flow_block_binder_type binder_type)
1397 {
1398 	struct tcf_block_owner_item *item;
1399 
1400 	item = kmalloc(sizeof(*item), GFP_KERNEL);
1401 	if (!item)
1402 		return -ENOMEM;
1403 	item->q = q;
1404 	item->binder_type = binder_type;
1405 	list_add(&item->list, &block->owner_list);
1406 	return 0;
1407 }
1408 
1409 static void tcf_block_owner_del(struct tcf_block *block,
1410 				struct Qdisc *q,
1411 				enum flow_block_binder_type binder_type)
1412 {
1413 	struct tcf_block_owner_item *item;
1414 
1415 	list_for_each_entry(item, &block->owner_list, list) {
1416 		if (item->q == q && item->binder_type == binder_type) {
1417 			list_del(&item->list);
1418 			kfree(item);
1419 			return;
1420 		}
1421 	}
1422 	WARN_ON(1);
1423 }
1424 
1425 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
1426 		      struct tcf_block_ext_info *ei,
1427 		      struct netlink_ext_ack *extack)
1428 {
1429 	struct net *net = qdisc_net(q);
1430 	struct tcf_block *block = NULL;
1431 	int err;
1432 
1433 	if (ei->block_index)
1434 		/* block_index not 0 means the shared block is requested */
1435 		block = tcf_block_refcnt_get(net, ei->block_index);
1436 
1437 	if (!block) {
1438 		block = tcf_block_create(net, q, ei->block_index, extack);
1439 		if (IS_ERR(block))
1440 			return PTR_ERR(block);
1441 		if (tcf_block_shared(block)) {
1442 			err = tcf_block_insert(block, net, extack);
1443 			if (err)
1444 				goto err_block_insert;
1445 		}
1446 	}
1447 
1448 	err = tcf_block_owner_add(block, q, ei->binder_type);
1449 	if (err)
1450 		goto err_block_owner_add;
1451 
1452 	tcf_block_owner_netif_keep_dst(block, q, ei->binder_type);
1453 
1454 	err = tcf_chain0_head_change_cb_add(block, ei, extack);
1455 	if (err)
1456 		goto err_chain0_head_change_cb_add;
1457 
1458 	err = tcf_block_offload_bind(block, q, ei, extack);
1459 	if (err)
1460 		goto err_block_offload_bind;
1461 
1462 	*p_block = block;
1463 	return 0;
1464 
1465 err_block_offload_bind:
1466 	tcf_chain0_head_change_cb_del(block, ei);
1467 err_chain0_head_change_cb_add:
1468 	tcf_block_owner_del(block, q, ei->binder_type);
1469 err_block_owner_add:
1470 err_block_insert:
1471 	tcf_block_refcnt_put(block, true);
1472 	return err;
1473 }
1474 EXPORT_SYMBOL(tcf_block_get_ext);
1475 
1476 static void tcf_chain_head_change_dflt(struct tcf_proto *tp_head, void *priv)
1477 {
1478 	struct tcf_proto __rcu **p_filter_chain = priv;
1479 
1480 	rcu_assign_pointer(*p_filter_chain, tp_head);
1481 }
1482 
1483 int tcf_block_get(struct tcf_block **p_block,
1484 		  struct tcf_proto __rcu **p_filter_chain, struct Qdisc *q,
1485 		  struct netlink_ext_ack *extack)
1486 {
1487 	struct tcf_block_ext_info ei = {
1488 		.chain_head_change = tcf_chain_head_change_dflt,
1489 		.chain_head_change_priv = p_filter_chain,
1490 	};
1491 
1492 	WARN_ON(!p_filter_chain);
1493 	return tcf_block_get_ext(p_block, q, &ei, extack);
1494 }
1495 EXPORT_SYMBOL(tcf_block_get);
1496 
1497 /* XXX: Standalone actions are not allowed to jump to any chain, and bound
1498  * actions should be all removed after flushing.
1499  */
1500 void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
1501 		       struct tcf_block_ext_info *ei)
1502 {
1503 	if (!block)
1504 		return;
1505 	tcf_chain0_head_change_cb_del(block, ei);
1506 	tcf_block_owner_del(block, q, ei->binder_type);
1507 
1508 	__tcf_block_put(block, q, ei, true);
1509 }
1510 EXPORT_SYMBOL(tcf_block_put_ext);
1511 
1512 void tcf_block_put(struct tcf_block *block)
1513 {
1514 	struct tcf_block_ext_info ei = {0, };
1515 
1516 	if (!block)
1517 		return;
1518 	tcf_block_put_ext(block, block->q, &ei);
1519 }
1520 
1521 EXPORT_SYMBOL(tcf_block_put);
1522 
1523 static int
1524 tcf_block_playback_offloads(struct tcf_block *block, flow_setup_cb_t *cb,
1525 			    void *cb_priv, bool add, bool offload_in_use,
1526 			    struct netlink_ext_ack *extack)
1527 {
1528 	struct tcf_chain *chain, *chain_prev;
1529 	struct tcf_proto *tp, *tp_prev;
1530 	int err;
1531 
1532 	lockdep_assert_held(&block->cb_lock);
1533 
1534 	for (chain = __tcf_get_next_chain(block, NULL);
1535 	     chain;
1536 	     chain_prev = chain,
1537 		     chain = __tcf_get_next_chain(block, chain),
1538 		     tcf_chain_put(chain_prev)) {
1539 		for (tp = __tcf_get_next_proto(chain, NULL); tp;
1540 		     tp_prev = tp,
1541 			     tp = __tcf_get_next_proto(chain, tp),
1542 			     tcf_proto_put(tp_prev, true, NULL)) {
1543 			if (tp->ops->reoffload) {
1544 				err = tp->ops->reoffload(tp, add, cb, cb_priv,
1545 							 extack);
1546 				if (err && add)
1547 					goto err_playback_remove;
1548 			} else if (add && offload_in_use) {
1549 				err = -EOPNOTSUPP;
1550 				NL_SET_ERR_MSG(extack, "Filter HW offload failed - classifier without re-offloading support");
1551 				goto err_playback_remove;
1552 			}
1553 		}
1554 	}
1555 
1556 	return 0;
1557 
1558 err_playback_remove:
1559 	tcf_proto_put(tp, true, NULL);
1560 	tcf_chain_put(chain);
1561 	tcf_block_playback_offloads(block, cb, cb_priv, false, offload_in_use,
1562 				    extack);
1563 	return err;
1564 }
1565 
1566 static int tcf_block_bind(struct tcf_block *block,
1567 			  struct flow_block_offload *bo)
1568 {
1569 	struct flow_block_cb *block_cb, *next;
1570 	int err, i = 0;
1571 
1572 	lockdep_assert_held(&block->cb_lock);
1573 
1574 	list_for_each_entry(block_cb, &bo->cb_list, list) {
1575 		err = tcf_block_playback_offloads(block, block_cb->cb,
1576 						  block_cb->cb_priv, true,
1577 						  tcf_block_offload_in_use(block),
1578 						  bo->extack);
1579 		if (err)
1580 			goto err_unroll;
1581 		if (!bo->unlocked_driver_cb)
1582 			block->lockeddevcnt++;
1583 
1584 		i++;
1585 	}
1586 	list_splice(&bo->cb_list, &block->flow_block.cb_list);
1587 
1588 	return 0;
1589 
1590 err_unroll:
1591 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1592 		list_del(&block_cb->driver_list);
1593 		if (i-- > 0) {
1594 			list_del(&block_cb->list);
1595 			tcf_block_playback_offloads(block, block_cb->cb,
1596 						    block_cb->cb_priv, false,
1597 						    tcf_block_offload_in_use(block),
1598 						    NULL);
1599 			if (!bo->unlocked_driver_cb)
1600 				block->lockeddevcnt--;
1601 		}
1602 		flow_block_cb_free(block_cb);
1603 	}
1604 
1605 	return err;
1606 }
1607 
1608 static void tcf_block_unbind(struct tcf_block *block,
1609 			     struct flow_block_offload *bo)
1610 {
1611 	struct flow_block_cb *block_cb, *next;
1612 
1613 	lockdep_assert_held(&block->cb_lock);
1614 
1615 	list_for_each_entry_safe(block_cb, next, &bo->cb_list, list) {
1616 		tcf_block_playback_offloads(block, block_cb->cb,
1617 					    block_cb->cb_priv, false,
1618 					    tcf_block_offload_in_use(block),
1619 					    NULL);
1620 		list_del(&block_cb->list);
1621 		flow_block_cb_free(block_cb);
1622 		if (!bo->unlocked_driver_cb)
1623 			block->lockeddevcnt--;
1624 	}
1625 }
1626 
1627 static int tcf_block_setup(struct tcf_block *block,
1628 			   struct flow_block_offload *bo)
1629 {
1630 	int err;
1631 
1632 	switch (bo->command) {
1633 	case FLOW_BLOCK_BIND:
1634 		err = tcf_block_bind(block, bo);
1635 		break;
1636 	case FLOW_BLOCK_UNBIND:
1637 		err = 0;
1638 		tcf_block_unbind(block, bo);
1639 		break;
1640 	default:
1641 		WARN_ON_ONCE(1);
1642 		err = -EOPNOTSUPP;
1643 	}
1644 
1645 	return err;
1646 }
1647 
1648 /* Main classifier routine: scans classifier chain attached
1649  * to this qdisc, (optionally) tests for protocol and asks
1650  * specific classifiers.
1651  */
1652 static inline int __tcf_classify(struct sk_buff *skb,
1653 				 const struct tcf_proto *tp,
1654 				 const struct tcf_proto *orig_tp,
1655 				 struct tcf_result *res,
1656 				 bool compat_mode,
1657 				 struct tcf_exts_miss_cookie_node *n,
1658 				 int act_index,
1659 				 u32 *last_executed_chain)
1660 {
1661 #ifdef CONFIG_NET_CLS_ACT
1662 	const int max_reclassify_loop = 16;
1663 	const struct tcf_proto *first_tp;
1664 	int limit = 0;
1665 
1666 reclassify:
1667 #endif
1668 	for (; tp; tp = rcu_dereference_bh(tp->next)) {
1669 		__be16 protocol = skb_protocol(skb, false);
1670 		int err = 0;
1671 
1672 		if (n) {
1673 			struct tcf_exts *exts;
1674 
1675 			if (n->tp_prio != tp->prio)
1676 				continue;
1677 
1678 			/* We re-lookup the tp and chain based on index instead
1679 			 * of having hard refs and locks to them, so do a sanity
1680 			 * check if any of tp,chain,exts was replaced by the
1681 			 * time we got here with a cookie from hardware.
1682 			 */
1683 			if (unlikely(n->tp != tp || n->tp->chain != n->chain ||
1684 				     !tp->ops->get_exts))
1685 				return TC_ACT_SHOT;
1686 
1687 			exts = tp->ops->get_exts(tp, n->handle);
1688 			if (unlikely(!exts || n->exts != exts))
1689 				return TC_ACT_SHOT;
1690 
1691 			n = NULL;
1692 			err = tcf_exts_exec_ex(skb, exts, act_index, res);
1693 		} else {
1694 			if (tp->protocol != protocol &&
1695 			    tp->protocol != htons(ETH_P_ALL))
1696 				continue;
1697 
1698 			err = tc_classify(skb, tp, res);
1699 		}
1700 #ifdef CONFIG_NET_CLS_ACT
1701 		if (unlikely(err == TC_ACT_RECLASSIFY && !compat_mode)) {
1702 			first_tp = orig_tp;
1703 			*last_executed_chain = first_tp->chain->index;
1704 			goto reset;
1705 		} else if (unlikely(TC_ACT_EXT_CMP(err, TC_ACT_GOTO_CHAIN))) {
1706 			first_tp = res->goto_tp;
1707 			*last_executed_chain = err & TC_ACT_EXT_VAL_MASK;
1708 			goto reset;
1709 		}
1710 #endif
1711 		if (err >= 0)
1712 			return err;
1713 	}
1714 
1715 	if (unlikely(n))
1716 		return TC_ACT_SHOT;
1717 
1718 	return TC_ACT_UNSPEC; /* signal: continue lookup */
1719 #ifdef CONFIG_NET_CLS_ACT
1720 reset:
1721 	if (unlikely(limit++ >= max_reclassify_loop)) {
1722 		net_notice_ratelimited("%u: reclassify loop, rule prio %u, protocol %02x\n",
1723 				       tp->chain->block->index,
1724 				       tp->prio & 0xffff,
1725 				       ntohs(tp->protocol));
1726 		return TC_ACT_SHOT;
1727 	}
1728 
1729 	tp = first_tp;
1730 	goto reclassify;
1731 #endif
1732 }
1733 
1734 int tcf_classify(struct sk_buff *skb,
1735 		 const struct tcf_block *block,
1736 		 const struct tcf_proto *tp,
1737 		 struct tcf_result *res, bool compat_mode)
1738 {
1739 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1740 	u32 last_executed_chain = 0;
1741 
1742 	return __tcf_classify(skb, tp, tp, res, compat_mode, NULL, 0,
1743 			      &last_executed_chain);
1744 #else
1745 	u32 last_executed_chain = tp ? tp->chain->index : 0;
1746 	struct tcf_exts_miss_cookie_node *n = NULL;
1747 	const struct tcf_proto *orig_tp = tp;
1748 	struct tc_skb_ext *ext;
1749 	int act_index = 0;
1750 	int ret;
1751 
1752 	if (block) {
1753 		ext = skb_ext_find(skb, TC_SKB_EXT);
1754 
1755 		if (ext && (ext->chain || ext->act_miss)) {
1756 			struct tcf_chain *fchain;
1757 			u32 chain;
1758 
1759 			if (ext->act_miss) {
1760 				n = tcf_exts_miss_cookie_lookup(ext->act_miss_cookie,
1761 								&act_index);
1762 				if (!n)
1763 					return TC_ACT_SHOT;
1764 
1765 				chain = n->chain_index;
1766 			} else {
1767 				chain = ext->chain;
1768 			}
1769 
1770 			fchain = tcf_chain_lookup_rcu(block, chain);
1771 			if (!fchain)
1772 				return TC_ACT_SHOT;
1773 
1774 			/* Consume, so cloned/redirect skbs won't inherit ext */
1775 			skb_ext_del(skb, TC_SKB_EXT);
1776 
1777 			tp = rcu_dereference_bh(fchain->filter_chain);
1778 			last_executed_chain = fchain->index;
1779 		}
1780 	}
1781 
1782 	ret = __tcf_classify(skb, tp, orig_tp, res, compat_mode, n, act_index,
1783 			     &last_executed_chain);
1784 
1785 	if (tc_skb_ext_tc_enabled()) {
1786 		/* If we missed on some chain */
1787 		if (ret == TC_ACT_UNSPEC && last_executed_chain) {
1788 			struct tc_skb_cb *cb = tc_skb_cb(skb);
1789 
1790 			ext = tc_skb_ext_alloc(skb);
1791 			if (WARN_ON_ONCE(!ext))
1792 				return TC_ACT_SHOT;
1793 			ext->chain = last_executed_chain;
1794 			ext->mru = cb->mru;
1795 			ext->post_ct = cb->post_ct;
1796 			ext->post_ct_snat = cb->post_ct_snat;
1797 			ext->post_ct_dnat = cb->post_ct_dnat;
1798 			ext->zone = cb->zone;
1799 		}
1800 	}
1801 
1802 	return ret;
1803 #endif
1804 }
1805 EXPORT_SYMBOL(tcf_classify);
1806 
1807 struct tcf_chain_info {
1808 	struct tcf_proto __rcu **pprev;
1809 	struct tcf_proto __rcu *next;
1810 };
1811 
1812 static struct tcf_proto *tcf_chain_tp_prev(struct tcf_chain *chain,
1813 					   struct tcf_chain_info *chain_info)
1814 {
1815 	return tcf_chain_dereference(*chain_info->pprev, chain);
1816 }
1817 
1818 static int tcf_chain_tp_insert(struct tcf_chain *chain,
1819 			       struct tcf_chain_info *chain_info,
1820 			       struct tcf_proto *tp)
1821 {
1822 	if (chain->flushing)
1823 		return -EAGAIN;
1824 
1825 	RCU_INIT_POINTER(tp->next, tcf_chain_tp_prev(chain, chain_info));
1826 	if (*chain_info->pprev == chain->filter_chain)
1827 		tcf_chain0_head_change(chain, tp);
1828 	tcf_proto_get(tp);
1829 	rcu_assign_pointer(*chain_info->pprev, tp);
1830 
1831 	return 0;
1832 }
1833 
1834 static void tcf_chain_tp_remove(struct tcf_chain *chain,
1835 				struct tcf_chain_info *chain_info,
1836 				struct tcf_proto *tp)
1837 {
1838 	struct tcf_proto *next = tcf_chain_dereference(chain_info->next, chain);
1839 
1840 	tcf_proto_mark_delete(tp);
1841 	if (tp == chain->filter_chain)
1842 		tcf_chain0_head_change(chain, next);
1843 	RCU_INIT_POINTER(*chain_info->pprev, next);
1844 }
1845 
1846 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1847 					   struct tcf_chain_info *chain_info,
1848 					   u32 protocol, u32 prio,
1849 					   bool prio_allocate);
1850 
1851 /* Try to insert new proto.
1852  * If proto with specified priority already exists, free new proto
1853  * and return existing one.
1854  */
1855 
1856 static struct tcf_proto *tcf_chain_tp_insert_unique(struct tcf_chain *chain,
1857 						    struct tcf_proto *tp_new,
1858 						    u32 protocol, u32 prio,
1859 						    bool rtnl_held)
1860 {
1861 	struct tcf_chain_info chain_info;
1862 	struct tcf_proto *tp;
1863 	int err = 0;
1864 
1865 	mutex_lock(&chain->filter_chain_lock);
1866 
1867 	if (tcf_proto_exists_destroying(chain, tp_new)) {
1868 		mutex_unlock(&chain->filter_chain_lock);
1869 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1870 		return ERR_PTR(-EAGAIN);
1871 	}
1872 
1873 	tp = tcf_chain_tp_find(chain, &chain_info,
1874 			       protocol, prio, false);
1875 	if (!tp)
1876 		err = tcf_chain_tp_insert(chain, &chain_info, tp_new);
1877 	mutex_unlock(&chain->filter_chain_lock);
1878 
1879 	if (tp) {
1880 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1881 		tp_new = tp;
1882 	} else if (err) {
1883 		tcf_proto_destroy(tp_new, rtnl_held, false, NULL);
1884 		tp_new = ERR_PTR(err);
1885 	}
1886 
1887 	return tp_new;
1888 }
1889 
1890 static void tcf_chain_tp_delete_empty(struct tcf_chain *chain,
1891 				      struct tcf_proto *tp, bool rtnl_held,
1892 				      struct netlink_ext_ack *extack)
1893 {
1894 	struct tcf_chain_info chain_info;
1895 	struct tcf_proto *tp_iter;
1896 	struct tcf_proto **pprev;
1897 	struct tcf_proto *next;
1898 
1899 	mutex_lock(&chain->filter_chain_lock);
1900 
1901 	/* Atomically find and remove tp from chain. */
1902 	for (pprev = &chain->filter_chain;
1903 	     (tp_iter = tcf_chain_dereference(*pprev, chain));
1904 	     pprev = &tp_iter->next) {
1905 		if (tp_iter == tp) {
1906 			chain_info.pprev = pprev;
1907 			chain_info.next = tp_iter->next;
1908 			WARN_ON(tp_iter->deleting);
1909 			break;
1910 		}
1911 	}
1912 	/* Verify that tp still exists and no new filters were inserted
1913 	 * concurrently.
1914 	 * Mark tp for deletion if it is empty.
1915 	 */
1916 	if (!tp_iter || !tcf_proto_check_delete(tp)) {
1917 		mutex_unlock(&chain->filter_chain_lock);
1918 		return;
1919 	}
1920 
1921 	tcf_proto_signal_destroying(chain, tp);
1922 	next = tcf_chain_dereference(chain_info.next, chain);
1923 	if (tp == chain->filter_chain)
1924 		tcf_chain0_head_change(chain, next);
1925 	RCU_INIT_POINTER(*chain_info.pprev, next);
1926 	mutex_unlock(&chain->filter_chain_lock);
1927 
1928 	tcf_proto_put(tp, rtnl_held, extack);
1929 }
1930 
1931 static struct tcf_proto *tcf_chain_tp_find(struct tcf_chain *chain,
1932 					   struct tcf_chain_info *chain_info,
1933 					   u32 protocol, u32 prio,
1934 					   bool prio_allocate)
1935 {
1936 	struct tcf_proto **pprev;
1937 	struct tcf_proto *tp;
1938 
1939 	/* Check the chain for existence of proto-tcf with this priority */
1940 	for (pprev = &chain->filter_chain;
1941 	     (tp = tcf_chain_dereference(*pprev, chain));
1942 	     pprev = &tp->next) {
1943 		if (tp->prio >= prio) {
1944 			if (tp->prio == prio) {
1945 				if (prio_allocate ||
1946 				    (tp->protocol != protocol && protocol))
1947 					return ERR_PTR(-EINVAL);
1948 			} else {
1949 				tp = NULL;
1950 			}
1951 			break;
1952 		}
1953 	}
1954 	chain_info->pprev = pprev;
1955 	if (tp) {
1956 		chain_info->next = tp->next;
1957 		tcf_proto_get(tp);
1958 	} else {
1959 		chain_info->next = NULL;
1960 	}
1961 	return tp;
1962 }
1963 
1964 static int tcf_fill_node(struct net *net, struct sk_buff *skb,
1965 			 struct tcf_proto *tp, struct tcf_block *block,
1966 			 struct Qdisc *q, u32 parent, void *fh,
1967 			 u32 portid, u32 seq, u16 flags, int event,
1968 			 bool terse_dump, bool rtnl_held,
1969 			 struct netlink_ext_ack *extack)
1970 {
1971 	struct tcmsg *tcm;
1972 	struct nlmsghdr  *nlh;
1973 	unsigned char *b = skb_tail_pointer(skb);
1974 
1975 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
1976 	if (!nlh)
1977 		goto out_nlmsg_trim;
1978 	tcm = nlmsg_data(nlh);
1979 	tcm->tcm_family = AF_UNSPEC;
1980 	tcm->tcm__pad1 = 0;
1981 	tcm->tcm__pad2 = 0;
1982 	if (q) {
1983 		tcm->tcm_ifindex = qdisc_dev(q)->ifindex;
1984 		tcm->tcm_parent = parent;
1985 	} else {
1986 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
1987 		tcm->tcm_block_index = block->index;
1988 	}
1989 	tcm->tcm_info = TC_H_MAKE(tp->prio, tp->protocol);
1990 	if (nla_put_string(skb, TCA_KIND, tp->ops->kind))
1991 		goto nla_put_failure;
1992 	if (nla_put_u32(skb, TCA_CHAIN, tp->chain->index))
1993 		goto nla_put_failure;
1994 	if (!fh) {
1995 		tcm->tcm_handle = 0;
1996 	} else if (terse_dump) {
1997 		if (tp->ops->terse_dump) {
1998 			if (tp->ops->terse_dump(net, tp, fh, skb, tcm,
1999 						rtnl_held) < 0)
2000 				goto nla_put_failure;
2001 		} else {
2002 			goto cls_op_not_supp;
2003 		}
2004 	} else {
2005 		if (tp->ops->dump &&
2006 		    tp->ops->dump(net, tp, fh, skb, tcm, rtnl_held) < 0)
2007 			goto nla_put_failure;
2008 	}
2009 
2010 	if (extack && extack->_msg &&
2011 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2012 		goto nla_put_failure;
2013 
2014 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2015 
2016 	return skb->len;
2017 
2018 out_nlmsg_trim:
2019 nla_put_failure:
2020 cls_op_not_supp:
2021 	nlmsg_trim(skb, b);
2022 	return -1;
2023 }
2024 
2025 static int tfilter_notify(struct net *net, struct sk_buff *oskb,
2026 			  struct nlmsghdr *n, struct tcf_proto *tp,
2027 			  struct tcf_block *block, struct Qdisc *q,
2028 			  u32 parent, void *fh, int event, bool unicast,
2029 			  bool rtnl_held, struct netlink_ext_ack *extack)
2030 {
2031 	struct sk_buff *skb;
2032 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2033 	int err = 0;
2034 
2035 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2036 	if (!skb)
2037 		return -ENOBUFS;
2038 
2039 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2040 			  n->nlmsg_seq, n->nlmsg_flags, event,
2041 			  false, rtnl_held, extack) <= 0) {
2042 		kfree_skb(skb);
2043 		return -EINVAL;
2044 	}
2045 
2046 	if (unicast)
2047 		err = rtnl_unicast(skb, net, portid);
2048 	else
2049 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2050 				     n->nlmsg_flags & NLM_F_ECHO);
2051 	return err;
2052 }
2053 
2054 static int tfilter_del_notify(struct net *net, struct sk_buff *oskb,
2055 			      struct nlmsghdr *n, struct tcf_proto *tp,
2056 			      struct tcf_block *block, struct Qdisc *q,
2057 			      u32 parent, void *fh, bool unicast, bool *last,
2058 			      bool rtnl_held, struct netlink_ext_ack *extack)
2059 {
2060 	struct sk_buff *skb;
2061 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2062 	int err;
2063 
2064 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2065 	if (!skb)
2066 		return -ENOBUFS;
2067 
2068 	if (tcf_fill_node(net, skb, tp, block, q, parent, fh, portid,
2069 			  n->nlmsg_seq, n->nlmsg_flags, RTM_DELTFILTER,
2070 			  false, rtnl_held, extack) <= 0) {
2071 		NL_SET_ERR_MSG(extack, "Failed to build del event notification");
2072 		kfree_skb(skb);
2073 		return -EINVAL;
2074 	}
2075 
2076 	err = tp->ops->delete(tp, fh, last, rtnl_held, extack);
2077 	if (err) {
2078 		kfree_skb(skb);
2079 		return err;
2080 	}
2081 
2082 	if (unicast)
2083 		err = rtnl_unicast(skb, net, portid);
2084 	else
2085 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2086 				     n->nlmsg_flags & NLM_F_ECHO);
2087 	if (err < 0)
2088 		NL_SET_ERR_MSG(extack, "Failed to send filter delete notification");
2089 
2090 	return err;
2091 }
2092 
2093 static void tfilter_notify_chain(struct net *net, struct sk_buff *oskb,
2094 				 struct tcf_block *block, struct Qdisc *q,
2095 				 u32 parent, struct nlmsghdr *n,
2096 				 struct tcf_chain *chain, int event,
2097 				 struct netlink_ext_ack *extack)
2098 {
2099 	struct tcf_proto *tp;
2100 
2101 	for (tp = tcf_get_next_proto(chain, NULL);
2102 	     tp; tp = tcf_get_next_proto(chain, tp))
2103 		tfilter_notify(net, oskb, n, tp, block, q, parent, NULL,
2104 			       event, false, true, extack);
2105 }
2106 
2107 static void tfilter_put(struct tcf_proto *tp, void *fh)
2108 {
2109 	if (tp->ops->put && fh)
2110 		tp->ops->put(tp, fh);
2111 }
2112 
2113 static bool is_qdisc_ingress(__u32 classid)
2114 {
2115 	return (TC_H_MIN(classid) == TC_H_MIN(TC_H_MIN_INGRESS));
2116 }
2117 
2118 static int tc_new_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2119 			  struct netlink_ext_ack *extack)
2120 {
2121 	struct net *net = sock_net(skb->sk);
2122 	struct nlattr *tca[TCA_MAX + 1];
2123 	char name[IFNAMSIZ];
2124 	struct tcmsg *t;
2125 	u32 protocol;
2126 	u32 prio;
2127 	bool prio_allocate;
2128 	u32 parent;
2129 	u32 chain_index;
2130 	struct Qdisc *q;
2131 	struct tcf_chain_info chain_info;
2132 	struct tcf_chain *chain;
2133 	struct tcf_block *block;
2134 	struct tcf_proto *tp;
2135 	unsigned long cl;
2136 	void *fh;
2137 	int err;
2138 	int tp_created;
2139 	bool rtnl_held = false;
2140 	u32 flags;
2141 
2142 replay:
2143 	tp_created = 0;
2144 
2145 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2146 				     rtm_tca_policy, extack);
2147 	if (err < 0)
2148 		return err;
2149 
2150 	t = nlmsg_data(n);
2151 	protocol = TC_H_MIN(t->tcm_info);
2152 	prio = TC_H_MAJ(t->tcm_info);
2153 	prio_allocate = false;
2154 	parent = t->tcm_parent;
2155 	tp = NULL;
2156 	cl = 0;
2157 	block = NULL;
2158 	q = NULL;
2159 	chain = NULL;
2160 	flags = 0;
2161 
2162 	if (prio == 0) {
2163 		/* If no priority is provided by the user,
2164 		 * we allocate one.
2165 		 */
2166 		if (n->nlmsg_flags & NLM_F_CREATE) {
2167 			prio = TC_H_MAKE(0x80000000U, 0U);
2168 			prio_allocate = true;
2169 		} else {
2170 			NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2171 			return -ENOENT;
2172 		}
2173 	}
2174 
2175 	/* Find head of filter chain. */
2176 
2177 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2178 	if (err)
2179 		return err;
2180 
2181 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2182 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2183 		err = -EINVAL;
2184 		goto errout;
2185 	}
2186 
2187 	/* Take rtnl mutex if rtnl_held was set to true on previous iteration,
2188 	 * block is shared (no qdisc found), qdisc is not unlocked, classifier
2189 	 * type is not specified, classifier is not unlocked.
2190 	 */
2191 	if (rtnl_held ||
2192 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2193 	    !tcf_proto_is_unlocked(name)) {
2194 		rtnl_held = true;
2195 		rtnl_lock();
2196 	}
2197 
2198 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2199 	if (err)
2200 		goto errout;
2201 
2202 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2203 				 extack);
2204 	if (IS_ERR(block)) {
2205 		err = PTR_ERR(block);
2206 		goto errout;
2207 	}
2208 	block->classid = parent;
2209 
2210 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2211 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2212 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2213 		err = -EINVAL;
2214 		goto errout;
2215 	}
2216 	chain = tcf_chain_get(block, chain_index, true);
2217 	if (!chain) {
2218 		NL_SET_ERR_MSG(extack, "Cannot create specified filter chain");
2219 		err = -ENOMEM;
2220 		goto errout;
2221 	}
2222 
2223 	mutex_lock(&chain->filter_chain_lock);
2224 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2225 			       prio, prio_allocate);
2226 	if (IS_ERR(tp)) {
2227 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2228 		err = PTR_ERR(tp);
2229 		goto errout_locked;
2230 	}
2231 
2232 	if (tp == NULL) {
2233 		struct tcf_proto *tp_new = NULL;
2234 
2235 		if (chain->flushing) {
2236 			err = -EAGAIN;
2237 			goto errout_locked;
2238 		}
2239 
2240 		/* Proto-tcf does not exist, create new one */
2241 
2242 		if (tca[TCA_KIND] == NULL || !protocol) {
2243 			NL_SET_ERR_MSG(extack, "Filter kind and protocol must be specified");
2244 			err = -EINVAL;
2245 			goto errout_locked;
2246 		}
2247 
2248 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2249 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2250 			err = -ENOENT;
2251 			goto errout_locked;
2252 		}
2253 
2254 		if (prio_allocate)
2255 			prio = tcf_auto_prio(tcf_chain_tp_prev(chain,
2256 							       &chain_info));
2257 
2258 		mutex_unlock(&chain->filter_chain_lock);
2259 		tp_new = tcf_proto_create(name, protocol, prio, chain,
2260 					  rtnl_held, extack);
2261 		if (IS_ERR(tp_new)) {
2262 			err = PTR_ERR(tp_new);
2263 			goto errout_tp;
2264 		}
2265 
2266 		tp_created = 1;
2267 		tp = tcf_chain_tp_insert_unique(chain, tp_new, protocol, prio,
2268 						rtnl_held);
2269 		if (IS_ERR(tp)) {
2270 			err = PTR_ERR(tp);
2271 			goto errout_tp;
2272 		}
2273 	} else {
2274 		mutex_unlock(&chain->filter_chain_lock);
2275 	}
2276 
2277 	if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2278 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2279 		err = -EINVAL;
2280 		goto errout;
2281 	}
2282 
2283 	fh = tp->ops->get(tp, t->tcm_handle);
2284 
2285 	if (!fh) {
2286 		if (!(n->nlmsg_flags & NLM_F_CREATE)) {
2287 			NL_SET_ERR_MSG(extack, "Need both RTM_NEWTFILTER and NLM_F_CREATE to create a new filter");
2288 			err = -ENOENT;
2289 			goto errout;
2290 		}
2291 	} else if (n->nlmsg_flags & NLM_F_EXCL) {
2292 		tfilter_put(tp, fh);
2293 		NL_SET_ERR_MSG(extack, "Filter already exists");
2294 		err = -EEXIST;
2295 		goto errout;
2296 	}
2297 
2298 	if (chain->tmplt_ops && chain->tmplt_ops != tp->ops) {
2299 		tfilter_put(tp, fh);
2300 		NL_SET_ERR_MSG(extack, "Chain template is set to a different filter kind");
2301 		err = -EINVAL;
2302 		goto errout;
2303 	}
2304 
2305 	if (!(n->nlmsg_flags & NLM_F_CREATE))
2306 		flags |= TCA_ACT_FLAGS_REPLACE;
2307 	if (!rtnl_held)
2308 		flags |= TCA_ACT_FLAGS_NO_RTNL;
2309 	if (is_qdisc_ingress(parent))
2310 		flags |= TCA_ACT_FLAGS_AT_INGRESS;
2311 	err = tp->ops->change(net, skb, tp, cl, t->tcm_handle, tca, &fh,
2312 			      flags, extack);
2313 	if (err == 0) {
2314 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2315 			       RTM_NEWTFILTER, false, rtnl_held, extack);
2316 		tfilter_put(tp, fh);
2317 		/* q pointer is NULL for shared blocks */
2318 		if (q)
2319 			q->flags &= ~TCQ_F_CAN_BYPASS;
2320 	}
2321 
2322 errout:
2323 	if (err && tp_created)
2324 		tcf_chain_tp_delete_empty(chain, tp, rtnl_held, NULL);
2325 errout_tp:
2326 	if (chain) {
2327 		if (tp && !IS_ERR(tp))
2328 			tcf_proto_put(tp, rtnl_held, NULL);
2329 		if (!tp_created)
2330 			tcf_chain_put(chain);
2331 	}
2332 	tcf_block_release(q, block, rtnl_held);
2333 
2334 	if (rtnl_held)
2335 		rtnl_unlock();
2336 
2337 	if (err == -EAGAIN) {
2338 		/* Take rtnl lock in case EAGAIN is caused by concurrent flush
2339 		 * of target chain.
2340 		 */
2341 		rtnl_held = true;
2342 		/* Replay the request. */
2343 		goto replay;
2344 	}
2345 	return err;
2346 
2347 errout_locked:
2348 	mutex_unlock(&chain->filter_chain_lock);
2349 	goto errout;
2350 }
2351 
2352 static int tc_del_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2353 			  struct netlink_ext_ack *extack)
2354 {
2355 	struct net *net = sock_net(skb->sk);
2356 	struct nlattr *tca[TCA_MAX + 1];
2357 	char name[IFNAMSIZ];
2358 	struct tcmsg *t;
2359 	u32 protocol;
2360 	u32 prio;
2361 	u32 parent;
2362 	u32 chain_index;
2363 	struct Qdisc *q = NULL;
2364 	struct tcf_chain_info chain_info;
2365 	struct tcf_chain *chain = NULL;
2366 	struct tcf_block *block = NULL;
2367 	struct tcf_proto *tp = NULL;
2368 	unsigned long cl = 0;
2369 	void *fh = NULL;
2370 	int err;
2371 	bool rtnl_held = false;
2372 
2373 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2374 				     rtm_tca_policy, extack);
2375 	if (err < 0)
2376 		return err;
2377 
2378 	t = nlmsg_data(n);
2379 	protocol = TC_H_MIN(t->tcm_info);
2380 	prio = TC_H_MAJ(t->tcm_info);
2381 	parent = t->tcm_parent;
2382 
2383 	if (prio == 0 && (protocol || t->tcm_handle || tca[TCA_KIND])) {
2384 		NL_SET_ERR_MSG(extack, "Cannot flush filters with protocol, handle or kind set");
2385 		return -ENOENT;
2386 	}
2387 
2388 	/* Find head of filter chain. */
2389 
2390 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2391 	if (err)
2392 		return err;
2393 
2394 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2395 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2396 		err = -EINVAL;
2397 		goto errout;
2398 	}
2399 	/* Take rtnl mutex if flushing whole chain, block is shared (no qdisc
2400 	 * found), qdisc is not unlocked, classifier type is not specified,
2401 	 * classifier is not unlocked.
2402 	 */
2403 	if (!prio ||
2404 	    (q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2405 	    !tcf_proto_is_unlocked(name)) {
2406 		rtnl_held = true;
2407 		rtnl_lock();
2408 	}
2409 
2410 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2411 	if (err)
2412 		goto errout;
2413 
2414 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2415 				 extack);
2416 	if (IS_ERR(block)) {
2417 		err = PTR_ERR(block);
2418 		goto errout;
2419 	}
2420 
2421 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2422 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2423 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2424 		err = -EINVAL;
2425 		goto errout;
2426 	}
2427 	chain = tcf_chain_get(block, chain_index, false);
2428 	if (!chain) {
2429 		/* User requested flush on non-existent chain. Nothing to do,
2430 		 * so just return success.
2431 		 */
2432 		if (prio == 0) {
2433 			err = 0;
2434 			goto errout;
2435 		}
2436 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2437 		err = -ENOENT;
2438 		goto errout;
2439 	}
2440 
2441 	if (prio == 0) {
2442 		tfilter_notify_chain(net, skb, block, q, parent, n,
2443 				     chain, RTM_DELTFILTER, extack);
2444 		tcf_chain_flush(chain, rtnl_held);
2445 		err = 0;
2446 		goto errout;
2447 	}
2448 
2449 	mutex_lock(&chain->filter_chain_lock);
2450 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2451 			       prio, false);
2452 	if (!tp || IS_ERR(tp)) {
2453 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2454 		err = tp ? PTR_ERR(tp) : -ENOENT;
2455 		goto errout_locked;
2456 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2457 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2458 		err = -EINVAL;
2459 		goto errout_locked;
2460 	} else if (t->tcm_handle == 0) {
2461 		tcf_proto_signal_destroying(chain, tp);
2462 		tcf_chain_tp_remove(chain, &chain_info, tp);
2463 		mutex_unlock(&chain->filter_chain_lock);
2464 
2465 		tcf_proto_put(tp, rtnl_held, NULL);
2466 		tfilter_notify(net, skb, n, tp, block, q, parent, fh,
2467 			       RTM_DELTFILTER, false, rtnl_held, extack);
2468 		err = 0;
2469 		goto errout;
2470 	}
2471 	mutex_unlock(&chain->filter_chain_lock);
2472 
2473 	fh = tp->ops->get(tp, t->tcm_handle);
2474 
2475 	if (!fh) {
2476 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2477 		err = -ENOENT;
2478 	} else {
2479 		bool last;
2480 
2481 		err = tfilter_del_notify(net, skb, n, tp, block,
2482 					 q, parent, fh, false, &last,
2483 					 rtnl_held, extack);
2484 
2485 		if (err)
2486 			goto errout;
2487 		if (last)
2488 			tcf_chain_tp_delete_empty(chain, tp, rtnl_held, extack);
2489 	}
2490 
2491 errout:
2492 	if (chain) {
2493 		if (tp && !IS_ERR(tp))
2494 			tcf_proto_put(tp, rtnl_held, NULL);
2495 		tcf_chain_put(chain);
2496 	}
2497 	tcf_block_release(q, block, rtnl_held);
2498 
2499 	if (rtnl_held)
2500 		rtnl_unlock();
2501 
2502 	return err;
2503 
2504 errout_locked:
2505 	mutex_unlock(&chain->filter_chain_lock);
2506 	goto errout;
2507 }
2508 
2509 static int tc_get_tfilter(struct sk_buff *skb, struct nlmsghdr *n,
2510 			  struct netlink_ext_ack *extack)
2511 {
2512 	struct net *net = sock_net(skb->sk);
2513 	struct nlattr *tca[TCA_MAX + 1];
2514 	char name[IFNAMSIZ];
2515 	struct tcmsg *t;
2516 	u32 protocol;
2517 	u32 prio;
2518 	u32 parent;
2519 	u32 chain_index;
2520 	struct Qdisc *q = NULL;
2521 	struct tcf_chain_info chain_info;
2522 	struct tcf_chain *chain = NULL;
2523 	struct tcf_block *block = NULL;
2524 	struct tcf_proto *tp = NULL;
2525 	unsigned long cl = 0;
2526 	void *fh = NULL;
2527 	int err;
2528 	bool rtnl_held = false;
2529 
2530 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2531 				     rtm_tca_policy, extack);
2532 	if (err < 0)
2533 		return err;
2534 
2535 	t = nlmsg_data(n);
2536 	protocol = TC_H_MIN(t->tcm_info);
2537 	prio = TC_H_MAJ(t->tcm_info);
2538 	parent = t->tcm_parent;
2539 
2540 	if (prio == 0) {
2541 		NL_SET_ERR_MSG(extack, "Invalid filter command with priority of zero");
2542 		return -ENOENT;
2543 	}
2544 
2545 	/* Find head of filter chain. */
2546 
2547 	err = __tcf_qdisc_find(net, &q, &parent, t->tcm_ifindex, false, extack);
2548 	if (err)
2549 		return err;
2550 
2551 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2552 		NL_SET_ERR_MSG(extack, "Specified TC filter name too long");
2553 		err = -EINVAL;
2554 		goto errout;
2555 	}
2556 	/* Take rtnl mutex if block is shared (no qdisc found), qdisc is not
2557 	 * unlocked, classifier type is not specified, classifier is not
2558 	 * unlocked.
2559 	 */
2560 	if ((q && !(q->ops->cl_ops->flags & QDISC_CLASS_OPS_DOIT_UNLOCKED)) ||
2561 	    !tcf_proto_is_unlocked(name)) {
2562 		rtnl_held = true;
2563 		rtnl_lock();
2564 	}
2565 
2566 	err = __tcf_qdisc_cl_find(q, parent, &cl, t->tcm_ifindex, extack);
2567 	if (err)
2568 		goto errout;
2569 
2570 	block = __tcf_block_find(net, q, cl, t->tcm_ifindex, t->tcm_block_index,
2571 				 extack);
2572 	if (IS_ERR(block)) {
2573 		err = PTR_ERR(block);
2574 		goto errout;
2575 	}
2576 
2577 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
2578 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
2579 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
2580 		err = -EINVAL;
2581 		goto errout;
2582 	}
2583 	chain = tcf_chain_get(block, chain_index, false);
2584 	if (!chain) {
2585 		NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
2586 		err = -EINVAL;
2587 		goto errout;
2588 	}
2589 
2590 	mutex_lock(&chain->filter_chain_lock);
2591 	tp = tcf_chain_tp_find(chain, &chain_info, protocol,
2592 			       prio, false);
2593 	mutex_unlock(&chain->filter_chain_lock);
2594 	if (!tp || IS_ERR(tp)) {
2595 		NL_SET_ERR_MSG(extack, "Filter with specified priority/protocol not found");
2596 		err = tp ? PTR_ERR(tp) : -ENOENT;
2597 		goto errout;
2598 	} else if (tca[TCA_KIND] && nla_strcmp(tca[TCA_KIND], tp->ops->kind)) {
2599 		NL_SET_ERR_MSG(extack, "Specified filter kind does not match existing one");
2600 		err = -EINVAL;
2601 		goto errout;
2602 	}
2603 
2604 	fh = tp->ops->get(tp, t->tcm_handle);
2605 
2606 	if (!fh) {
2607 		NL_SET_ERR_MSG(extack, "Specified filter handle not found");
2608 		err = -ENOENT;
2609 	} else {
2610 		err = tfilter_notify(net, skb, n, tp, block, q, parent,
2611 				     fh, RTM_NEWTFILTER, true, rtnl_held, NULL);
2612 		if (err < 0)
2613 			NL_SET_ERR_MSG(extack, "Failed to send filter notify message");
2614 	}
2615 
2616 	tfilter_put(tp, fh);
2617 errout:
2618 	if (chain) {
2619 		if (tp && !IS_ERR(tp))
2620 			tcf_proto_put(tp, rtnl_held, NULL);
2621 		tcf_chain_put(chain);
2622 	}
2623 	tcf_block_release(q, block, rtnl_held);
2624 
2625 	if (rtnl_held)
2626 		rtnl_unlock();
2627 
2628 	return err;
2629 }
2630 
2631 struct tcf_dump_args {
2632 	struct tcf_walker w;
2633 	struct sk_buff *skb;
2634 	struct netlink_callback *cb;
2635 	struct tcf_block *block;
2636 	struct Qdisc *q;
2637 	u32 parent;
2638 	bool terse_dump;
2639 };
2640 
2641 static int tcf_node_dump(struct tcf_proto *tp, void *n, struct tcf_walker *arg)
2642 {
2643 	struct tcf_dump_args *a = (void *)arg;
2644 	struct net *net = sock_net(a->skb->sk);
2645 
2646 	return tcf_fill_node(net, a->skb, tp, a->block, a->q, a->parent,
2647 			     n, NETLINK_CB(a->cb->skb).portid,
2648 			     a->cb->nlh->nlmsg_seq, NLM_F_MULTI,
2649 			     RTM_NEWTFILTER, a->terse_dump, true, NULL);
2650 }
2651 
2652 static bool tcf_chain_dump(struct tcf_chain *chain, struct Qdisc *q, u32 parent,
2653 			   struct sk_buff *skb, struct netlink_callback *cb,
2654 			   long index_start, long *p_index, bool terse)
2655 {
2656 	struct net *net = sock_net(skb->sk);
2657 	struct tcf_block *block = chain->block;
2658 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2659 	struct tcf_proto *tp, *tp_prev;
2660 	struct tcf_dump_args arg;
2661 
2662 	for (tp = __tcf_get_next_proto(chain, NULL);
2663 	     tp;
2664 	     tp_prev = tp,
2665 		     tp = __tcf_get_next_proto(chain, tp),
2666 		     tcf_proto_put(tp_prev, true, NULL),
2667 		     (*p_index)++) {
2668 		if (*p_index < index_start)
2669 			continue;
2670 		if (TC_H_MAJ(tcm->tcm_info) &&
2671 		    TC_H_MAJ(tcm->tcm_info) != tp->prio)
2672 			continue;
2673 		if (TC_H_MIN(tcm->tcm_info) &&
2674 		    TC_H_MIN(tcm->tcm_info) != tp->protocol)
2675 			continue;
2676 		if (*p_index > index_start)
2677 			memset(&cb->args[1], 0,
2678 			       sizeof(cb->args) - sizeof(cb->args[0]));
2679 		if (cb->args[1] == 0) {
2680 			if (tcf_fill_node(net, skb, tp, block, q, parent, NULL,
2681 					  NETLINK_CB(cb->skb).portid,
2682 					  cb->nlh->nlmsg_seq, NLM_F_MULTI,
2683 					  RTM_NEWTFILTER, false, true, NULL) <= 0)
2684 				goto errout;
2685 			cb->args[1] = 1;
2686 		}
2687 		if (!tp->ops->walk)
2688 			continue;
2689 		arg.w.fn = tcf_node_dump;
2690 		arg.skb = skb;
2691 		arg.cb = cb;
2692 		arg.block = block;
2693 		arg.q = q;
2694 		arg.parent = parent;
2695 		arg.w.stop = 0;
2696 		arg.w.skip = cb->args[1] - 1;
2697 		arg.w.count = 0;
2698 		arg.w.cookie = cb->args[2];
2699 		arg.terse_dump = terse;
2700 		tp->ops->walk(tp, &arg.w, true);
2701 		cb->args[2] = arg.w.cookie;
2702 		cb->args[1] = arg.w.count + 1;
2703 		if (arg.w.stop)
2704 			goto errout;
2705 	}
2706 	return true;
2707 
2708 errout:
2709 	tcf_proto_put(tp, true, NULL);
2710 	return false;
2711 }
2712 
2713 static const struct nla_policy tcf_tfilter_dump_policy[TCA_MAX + 1] = {
2714 	[TCA_DUMP_FLAGS] = NLA_POLICY_BITFIELD32(TCA_DUMP_FLAGS_TERSE),
2715 };
2716 
2717 /* called with RTNL */
2718 static int tc_dump_tfilter(struct sk_buff *skb, struct netlink_callback *cb)
2719 {
2720 	struct tcf_chain *chain, *chain_prev;
2721 	struct net *net = sock_net(skb->sk);
2722 	struct nlattr *tca[TCA_MAX + 1];
2723 	struct Qdisc *q = NULL;
2724 	struct tcf_block *block;
2725 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
2726 	bool terse_dump = false;
2727 	long index_start;
2728 	long index;
2729 	u32 parent;
2730 	int err;
2731 
2732 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
2733 		return skb->len;
2734 
2735 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
2736 				     tcf_tfilter_dump_policy, cb->extack);
2737 	if (err)
2738 		return err;
2739 
2740 	if (tca[TCA_DUMP_FLAGS]) {
2741 		struct nla_bitfield32 flags =
2742 			nla_get_bitfield32(tca[TCA_DUMP_FLAGS]);
2743 
2744 		terse_dump = flags.value & TCA_DUMP_FLAGS_TERSE;
2745 	}
2746 
2747 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
2748 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
2749 		if (!block)
2750 			goto out;
2751 		/* If we work with block index, q is NULL and parent value
2752 		 * will never be used in the following code. The check
2753 		 * in tcf_fill_node prevents it. However, compiler does not
2754 		 * see that far, so set parent to zero to silence the warning
2755 		 * about parent being uninitialized.
2756 		 */
2757 		parent = 0;
2758 	} else {
2759 		const struct Qdisc_class_ops *cops;
2760 		struct net_device *dev;
2761 		unsigned long cl = 0;
2762 
2763 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
2764 		if (!dev)
2765 			return skb->len;
2766 
2767 		parent = tcm->tcm_parent;
2768 		if (!parent)
2769 			q = rtnl_dereference(dev->qdisc);
2770 		else
2771 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
2772 		if (!q)
2773 			goto out;
2774 		cops = q->ops->cl_ops;
2775 		if (!cops)
2776 			goto out;
2777 		if (!cops->tcf_block)
2778 			goto out;
2779 		if (TC_H_MIN(tcm->tcm_parent)) {
2780 			cl = cops->find(q, tcm->tcm_parent);
2781 			if (cl == 0)
2782 				goto out;
2783 		}
2784 		block = cops->tcf_block(q, cl, NULL);
2785 		if (!block)
2786 			goto out;
2787 		parent = block->classid;
2788 		if (tcf_block_shared(block))
2789 			q = NULL;
2790 	}
2791 
2792 	index_start = cb->args[0];
2793 	index = 0;
2794 
2795 	for (chain = __tcf_get_next_chain(block, NULL);
2796 	     chain;
2797 	     chain_prev = chain,
2798 		     chain = __tcf_get_next_chain(block, chain),
2799 		     tcf_chain_put(chain_prev)) {
2800 		if (tca[TCA_CHAIN] &&
2801 		    nla_get_u32(tca[TCA_CHAIN]) != chain->index)
2802 			continue;
2803 		if (!tcf_chain_dump(chain, q, parent, skb, cb,
2804 				    index_start, &index, terse_dump)) {
2805 			tcf_chain_put(chain);
2806 			err = -EMSGSIZE;
2807 			break;
2808 		}
2809 	}
2810 
2811 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
2812 		tcf_block_refcnt_put(block, true);
2813 	cb->args[0] = index;
2814 
2815 out:
2816 	/* If we did no progress, the error (EMSGSIZE) is real */
2817 	if (skb->len == 0 && err)
2818 		return err;
2819 	return skb->len;
2820 }
2821 
2822 static int tc_chain_fill_node(const struct tcf_proto_ops *tmplt_ops,
2823 			      void *tmplt_priv, u32 chain_index,
2824 			      struct net *net, struct sk_buff *skb,
2825 			      struct tcf_block *block,
2826 			      u32 portid, u32 seq, u16 flags, int event,
2827 			      struct netlink_ext_ack *extack)
2828 {
2829 	unsigned char *b = skb_tail_pointer(skb);
2830 	const struct tcf_proto_ops *ops;
2831 	struct nlmsghdr *nlh;
2832 	struct tcmsg *tcm;
2833 	void *priv;
2834 
2835 	ops = tmplt_ops;
2836 	priv = tmplt_priv;
2837 
2838 	nlh = nlmsg_put(skb, portid, seq, event, sizeof(*tcm), flags);
2839 	if (!nlh)
2840 		goto out_nlmsg_trim;
2841 	tcm = nlmsg_data(nlh);
2842 	tcm->tcm_family = AF_UNSPEC;
2843 	tcm->tcm__pad1 = 0;
2844 	tcm->tcm__pad2 = 0;
2845 	tcm->tcm_handle = 0;
2846 	if (block->q) {
2847 		tcm->tcm_ifindex = qdisc_dev(block->q)->ifindex;
2848 		tcm->tcm_parent = block->q->handle;
2849 	} else {
2850 		tcm->tcm_ifindex = TCM_IFINDEX_MAGIC_BLOCK;
2851 		tcm->tcm_block_index = block->index;
2852 	}
2853 
2854 	if (nla_put_u32(skb, TCA_CHAIN, chain_index))
2855 		goto nla_put_failure;
2856 
2857 	if (ops) {
2858 		if (nla_put_string(skb, TCA_KIND, ops->kind))
2859 			goto nla_put_failure;
2860 		if (ops->tmplt_dump(skb, net, priv) < 0)
2861 			goto nla_put_failure;
2862 	}
2863 
2864 	if (extack && extack->_msg &&
2865 	    nla_put_string(skb, TCA_EXT_WARN_MSG, extack->_msg))
2866 		goto out_nlmsg_trim;
2867 
2868 	nlh->nlmsg_len = skb_tail_pointer(skb) - b;
2869 
2870 	return skb->len;
2871 
2872 out_nlmsg_trim:
2873 nla_put_failure:
2874 	nlmsg_trim(skb, b);
2875 	return -EMSGSIZE;
2876 }
2877 
2878 static int tc_chain_notify(struct tcf_chain *chain, struct sk_buff *oskb,
2879 			   u32 seq, u16 flags, int event, bool unicast,
2880 			   struct netlink_ext_ack *extack)
2881 {
2882 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2883 	struct tcf_block *block = chain->block;
2884 	struct net *net = block->net;
2885 	struct sk_buff *skb;
2886 	int err = 0;
2887 
2888 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2889 	if (!skb)
2890 		return -ENOBUFS;
2891 
2892 	if (tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
2893 			       chain->index, net, skb, block, portid,
2894 			       seq, flags, event, extack) <= 0) {
2895 		kfree_skb(skb);
2896 		return -EINVAL;
2897 	}
2898 
2899 	if (unicast)
2900 		err = rtnl_unicast(skb, net, portid);
2901 	else
2902 		err = rtnetlink_send(skb, net, portid, RTNLGRP_TC,
2903 				     flags & NLM_F_ECHO);
2904 
2905 	return err;
2906 }
2907 
2908 static int tc_chain_notify_delete(const struct tcf_proto_ops *tmplt_ops,
2909 				  void *tmplt_priv, u32 chain_index,
2910 				  struct tcf_block *block, struct sk_buff *oskb,
2911 				  u32 seq, u16 flags, bool unicast)
2912 {
2913 	u32 portid = oskb ? NETLINK_CB(oskb).portid : 0;
2914 	struct net *net = block->net;
2915 	struct sk_buff *skb;
2916 
2917 	skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
2918 	if (!skb)
2919 		return -ENOBUFS;
2920 
2921 	if (tc_chain_fill_node(tmplt_ops, tmplt_priv, chain_index, net, skb,
2922 			       block, portid, seq, flags, RTM_DELCHAIN, NULL) <= 0) {
2923 		kfree_skb(skb);
2924 		return -EINVAL;
2925 	}
2926 
2927 	if (unicast)
2928 		return rtnl_unicast(skb, net, portid);
2929 
2930 	return rtnetlink_send(skb, net, portid, RTNLGRP_TC, flags & NLM_F_ECHO);
2931 }
2932 
2933 static int tc_chain_tmplt_add(struct tcf_chain *chain, struct net *net,
2934 			      struct nlattr **tca,
2935 			      struct netlink_ext_ack *extack)
2936 {
2937 	const struct tcf_proto_ops *ops;
2938 	char name[IFNAMSIZ];
2939 	void *tmplt_priv;
2940 
2941 	/* If kind is not set, user did not specify template. */
2942 	if (!tca[TCA_KIND])
2943 		return 0;
2944 
2945 	if (tcf_proto_check_kind(tca[TCA_KIND], name)) {
2946 		NL_SET_ERR_MSG(extack, "Specified TC chain template name too long");
2947 		return -EINVAL;
2948 	}
2949 
2950 	ops = tcf_proto_lookup_ops(name, true, extack);
2951 	if (IS_ERR(ops))
2952 		return PTR_ERR(ops);
2953 	if (!ops->tmplt_create || !ops->tmplt_destroy || !ops->tmplt_dump) {
2954 		NL_SET_ERR_MSG(extack, "Chain templates are not supported with specified classifier");
2955 		return -EOPNOTSUPP;
2956 	}
2957 
2958 	tmplt_priv = ops->tmplt_create(net, chain, tca, extack);
2959 	if (IS_ERR(tmplt_priv)) {
2960 		module_put(ops->owner);
2961 		return PTR_ERR(tmplt_priv);
2962 	}
2963 	chain->tmplt_ops = ops;
2964 	chain->tmplt_priv = tmplt_priv;
2965 	return 0;
2966 }
2967 
2968 static void tc_chain_tmplt_del(const struct tcf_proto_ops *tmplt_ops,
2969 			       void *tmplt_priv)
2970 {
2971 	/* If template ops are set, no work to do for us. */
2972 	if (!tmplt_ops)
2973 		return;
2974 
2975 	tmplt_ops->tmplt_destroy(tmplt_priv);
2976 	module_put(tmplt_ops->owner);
2977 }
2978 
2979 /* Add/delete/get a chain */
2980 
2981 static int tc_ctl_chain(struct sk_buff *skb, struct nlmsghdr *n,
2982 			struct netlink_ext_ack *extack)
2983 {
2984 	struct net *net = sock_net(skb->sk);
2985 	struct nlattr *tca[TCA_MAX + 1];
2986 	struct tcmsg *t;
2987 	u32 parent;
2988 	u32 chain_index;
2989 	struct Qdisc *q;
2990 	struct tcf_chain *chain;
2991 	struct tcf_block *block;
2992 	unsigned long cl;
2993 	int err;
2994 
2995 replay:
2996 	q = NULL;
2997 	err = nlmsg_parse_deprecated(n, sizeof(*t), tca, TCA_MAX,
2998 				     rtm_tca_policy, extack);
2999 	if (err < 0)
3000 		return err;
3001 
3002 	t = nlmsg_data(n);
3003 	parent = t->tcm_parent;
3004 	cl = 0;
3005 
3006 	block = tcf_block_find(net, &q, &parent, &cl,
3007 			       t->tcm_ifindex, t->tcm_block_index, extack);
3008 	if (IS_ERR(block))
3009 		return PTR_ERR(block);
3010 
3011 	chain_index = tca[TCA_CHAIN] ? nla_get_u32(tca[TCA_CHAIN]) : 0;
3012 	if (chain_index > TC_ACT_EXT_VAL_MASK) {
3013 		NL_SET_ERR_MSG(extack, "Specified chain index exceeds upper limit");
3014 		err = -EINVAL;
3015 		goto errout_block;
3016 	}
3017 
3018 	mutex_lock(&block->lock);
3019 	chain = tcf_chain_lookup(block, chain_index);
3020 	if (n->nlmsg_type == RTM_NEWCHAIN) {
3021 		if (chain) {
3022 			if (tcf_chain_held_by_acts_only(chain)) {
3023 				/* The chain exists only because there is
3024 				 * some action referencing it.
3025 				 */
3026 				tcf_chain_hold(chain);
3027 			} else {
3028 				NL_SET_ERR_MSG(extack, "Filter chain already exists");
3029 				err = -EEXIST;
3030 				goto errout_block_locked;
3031 			}
3032 		} else {
3033 			if (!(n->nlmsg_flags & NLM_F_CREATE)) {
3034 				NL_SET_ERR_MSG(extack, "Need both RTM_NEWCHAIN and NLM_F_CREATE to create a new chain");
3035 				err = -ENOENT;
3036 				goto errout_block_locked;
3037 			}
3038 			chain = tcf_chain_create(block, chain_index);
3039 			if (!chain) {
3040 				NL_SET_ERR_MSG(extack, "Failed to create filter chain");
3041 				err = -ENOMEM;
3042 				goto errout_block_locked;
3043 			}
3044 		}
3045 	} else {
3046 		if (!chain || tcf_chain_held_by_acts_only(chain)) {
3047 			NL_SET_ERR_MSG(extack, "Cannot find specified filter chain");
3048 			err = -EINVAL;
3049 			goto errout_block_locked;
3050 		}
3051 		tcf_chain_hold(chain);
3052 	}
3053 
3054 	if (n->nlmsg_type == RTM_NEWCHAIN) {
3055 		/* Modifying chain requires holding parent block lock. In case
3056 		 * the chain was successfully added, take a reference to the
3057 		 * chain. This ensures that an empty chain does not disappear at
3058 		 * the end of this function.
3059 		 */
3060 		tcf_chain_hold(chain);
3061 		chain->explicitly_created = true;
3062 	}
3063 	mutex_unlock(&block->lock);
3064 
3065 	switch (n->nlmsg_type) {
3066 	case RTM_NEWCHAIN:
3067 		err = tc_chain_tmplt_add(chain, net, tca, extack);
3068 		if (err) {
3069 			tcf_chain_put_explicitly_created(chain);
3070 			goto errout;
3071 		}
3072 
3073 		tc_chain_notify(chain, NULL, 0, NLM_F_CREATE | NLM_F_EXCL,
3074 				RTM_NEWCHAIN, false, extack);
3075 		break;
3076 	case RTM_DELCHAIN:
3077 		tfilter_notify_chain(net, skb, block, q, parent, n,
3078 				     chain, RTM_DELTFILTER, extack);
3079 		/* Flush the chain first as the user requested chain removal. */
3080 		tcf_chain_flush(chain, true);
3081 		/* In case the chain was successfully deleted, put a reference
3082 		 * to the chain previously taken during addition.
3083 		 */
3084 		tcf_chain_put_explicitly_created(chain);
3085 		break;
3086 	case RTM_GETCHAIN:
3087 		err = tc_chain_notify(chain, skb, n->nlmsg_seq,
3088 				      n->nlmsg_flags, n->nlmsg_type, true, extack);
3089 		if (err < 0)
3090 			NL_SET_ERR_MSG(extack, "Failed to send chain notify message");
3091 		break;
3092 	default:
3093 		err = -EOPNOTSUPP;
3094 		NL_SET_ERR_MSG(extack, "Unsupported message type");
3095 		goto errout;
3096 	}
3097 
3098 errout:
3099 	tcf_chain_put(chain);
3100 errout_block:
3101 	tcf_block_release(q, block, true);
3102 	if (err == -EAGAIN)
3103 		/* Replay the request. */
3104 		goto replay;
3105 	return err;
3106 
3107 errout_block_locked:
3108 	mutex_unlock(&block->lock);
3109 	goto errout_block;
3110 }
3111 
3112 /* called with RTNL */
3113 static int tc_dump_chain(struct sk_buff *skb, struct netlink_callback *cb)
3114 {
3115 	struct net *net = sock_net(skb->sk);
3116 	struct nlattr *tca[TCA_MAX + 1];
3117 	struct Qdisc *q = NULL;
3118 	struct tcf_block *block;
3119 	struct tcmsg *tcm = nlmsg_data(cb->nlh);
3120 	struct tcf_chain *chain;
3121 	long index_start;
3122 	long index;
3123 	int err;
3124 
3125 	if (nlmsg_len(cb->nlh) < sizeof(*tcm))
3126 		return skb->len;
3127 
3128 	err = nlmsg_parse_deprecated(cb->nlh, sizeof(*tcm), tca, TCA_MAX,
3129 				     rtm_tca_policy, cb->extack);
3130 	if (err)
3131 		return err;
3132 
3133 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK) {
3134 		block = tcf_block_refcnt_get(net, tcm->tcm_block_index);
3135 		if (!block)
3136 			goto out;
3137 	} else {
3138 		const struct Qdisc_class_ops *cops;
3139 		struct net_device *dev;
3140 		unsigned long cl = 0;
3141 
3142 		dev = __dev_get_by_index(net, tcm->tcm_ifindex);
3143 		if (!dev)
3144 			return skb->len;
3145 
3146 		if (!tcm->tcm_parent)
3147 			q = rtnl_dereference(dev->qdisc);
3148 		else
3149 			q = qdisc_lookup(dev, TC_H_MAJ(tcm->tcm_parent));
3150 
3151 		if (!q)
3152 			goto out;
3153 		cops = q->ops->cl_ops;
3154 		if (!cops)
3155 			goto out;
3156 		if (!cops->tcf_block)
3157 			goto out;
3158 		if (TC_H_MIN(tcm->tcm_parent)) {
3159 			cl = cops->find(q, tcm->tcm_parent);
3160 			if (cl == 0)
3161 				goto out;
3162 		}
3163 		block = cops->tcf_block(q, cl, NULL);
3164 		if (!block)
3165 			goto out;
3166 		if (tcf_block_shared(block))
3167 			q = NULL;
3168 	}
3169 
3170 	index_start = cb->args[0];
3171 	index = 0;
3172 
3173 	mutex_lock(&block->lock);
3174 	list_for_each_entry(chain, &block->chain_list, list) {
3175 		if ((tca[TCA_CHAIN] &&
3176 		     nla_get_u32(tca[TCA_CHAIN]) != chain->index))
3177 			continue;
3178 		if (index < index_start) {
3179 			index++;
3180 			continue;
3181 		}
3182 		if (tcf_chain_held_by_acts_only(chain))
3183 			continue;
3184 		err = tc_chain_fill_node(chain->tmplt_ops, chain->tmplt_priv,
3185 					 chain->index, net, skb, block,
3186 					 NETLINK_CB(cb->skb).portid,
3187 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
3188 					 RTM_NEWCHAIN, NULL);
3189 		if (err <= 0)
3190 			break;
3191 		index++;
3192 	}
3193 	mutex_unlock(&block->lock);
3194 
3195 	if (tcm->tcm_ifindex == TCM_IFINDEX_MAGIC_BLOCK)
3196 		tcf_block_refcnt_put(block, true);
3197 	cb->args[0] = index;
3198 
3199 out:
3200 	/* If we did no progress, the error (EMSGSIZE) is real */
3201 	if (skb->len == 0 && err)
3202 		return err;
3203 	return skb->len;
3204 }
3205 
3206 int tcf_exts_init_ex(struct tcf_exts *exts, struct net *net, int action,
3207 		     int police, struct tcf_proto *tp, u32 handle,
3208 		     bool use_action_miss)
3209 {
3210 	int err = 0;
3211 
3212 #ifdef CONFIG_NET_CLS_ACT
3213 	exts->type = 0;
3214 	exts->nr_actions = 0;
3215 	exts->miss_cookie_node = NULL;
3216 	/* Note: we do not own yet a reference on net.
3217 	 * This reference might be taken later from tcf_exts_get_net().
3218 	 */
3219 	exts->net = net;
3220 	exts->actions = kcalloc(TCA_ACT_MAX_PRIO, sizeof(struct tc_action *),
3221 				GFP_KERNEL);
3222 	if (!exts->actions)
3223 		return -ENOMEM;
3224 #endif
3225 
3226 	exts->action = action;
3227 	exts->police = police;
3228 
3229 	if (!use_action_miss)
3230 		return 0;
3231 
3232 	err = tcf_exts_miss_cookie_base_alloc(exts, tp, handle);
3233 	if (err)
3234 		goto err_miss_alloc;
3235 
3236 	return 0;
3237 
3238 err_miss_alloc:
3239 	tcf_exts_destroy(exts);
3240 #ifdef CONFIG_NET_CLS_ACT
3241 	exts->actions = NULL;
3242 #endif
3243 	return err;
3244 }
3245 EXPORT_SYMBOL(tcf_exts_init_ex);
3246 
3247 void tcf_exts_destroy(struct tcf_exts *exts)
3248 {
3249 	tcf_exts_miss_cookie_base_destroy(exts);
3250 
3251 #ifdef CONFIG_NET_CLS_ACT
3252 	if (exts->actions) {
3253 		tcf_action_destroy(exts->actions, TCA_ACT_UNBIND);
3254 		kfree(exts->actions);
3255 	}
3256 	exts->nr_actions = 0;
3257 #endif
3258 }
3259 EXPORT_SYMBOL(tcf_exts_destroy);
3260 
3261 int tcf_exts_validate_ex(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3262 			 struct nlattr *rate_tlv, struct tcf_exts *exts,
3263 			 u32 flags, u32 fl_flags, struct netlink_ext_ack *extack)
3264 {
3265 #ifdef CONFIG_NET_CLS_ACT
3266 	{
3267 		int init_res[TCA_ACT_MAX_PRIO] = {};
3268 		struct tc_action *act;
3269 		size_t attr_size = 0;
3270 
3271 		if (exts->police && tb[exts->police]) {
3272 			struct tc_action_ops *a_o;
3273 
3274 			a_o = tc_action_load_ops(tb[exts->police], true,
3275 						 !(flags & TCA_ACT_FLAGS_NO_RTNL),
3276 						 extack);
3277 			if (IS_ERR(a_o))
3278 				return PTR_ERR(a_o);
3279 			flags |= TCA_ACT_FLAGS_POLICE | TCA_ACT_FLAGS_BIND;
3280 			act = tcf_action_init_1(net, tp, tb[exts->police],
3281 						rate_tlv, a_o, init_res, flags,
3282 						extack);
3283 			module_put(a_o->owner);
3284 			if (IS_ERR(act))
3285 				return PTR_ERR(act);
3286 
3287 			act->type = exts->type = TCA_OLD_COMPAT;
3288 			exts->actions[0] = act;
3289 			exts->nr_actions = 1;
3290 			tcf_idr_insert_many(exts->actions);
3291 		} else if (exts->action && tb[exts->action]) {
3292 			int err;
3293 
3294 			flags |= TCA_ACT_FLAGS_BIND;
3295 			err = tcf_action_init(net, tp, tb[exts->action],
3296 					      rate_tlv, exts->actions, init_res,
3297 					      &attr_size, flags, fl_flags,
3298 					      extack);
3299 			if (err < 0)
3300 				return err;
3301 			exts->nr_actions = err;
3302 		}
3303 	}
3304 #else
3305 	if ((exts->action && tb[exts->action]) ||
3306 	    (exts->police && tb[exts->police])) {
3307 		NL_SET_ERR_MSG(extack, "Classifier actions are not supported per compile options (CONFIG_NET_CLS_ACT)");
3308 		return -EOPNOTSUPP;
3309 	}
3310 #endif
3311 
3312 	return 0;
3313 }
3314 EXPORT_SYMBOL(tcf_exts_validate_ex);
3315 
3316 int tcf_exts_validate(struct net *net, struct tcf_proto *tp, struct nlattr **tb,
3317 		      struct nlattr *rate_tlv, struct tcf_exts *exts,
3318 		      u32 flags, struct netlink_ext_ack *extack)
3319 {
3320 	return tcf_exts_validate_ex(net, tp, tb, rate_tlv, exts,
3321 				    flags, 0, extack);
3322 }
3323 EXPORT_SYMBOL(tcf_exts_validate);
3324 
3325 void tcf_exts_change(struct tcf_exts *dst, struct tcf_exts *src)
3326 {
3327 #ifdef CONFIG_NET_CLS_ACT
3328 	struct tcf_exts old = *dst;
3329 
3330 	*dst = *src;
3331 	tcf_exts_destroy(&old);
3332 #endif
3333 }
3334 EXPORT_SYMBOL(tcf_exts_change);
3335 
3336 #ifdef CONFIG_NET_CLS_ACT
3337 static struct tc_action *tcf_exts_first_act(struct tcf_exts *exts)
3338 {
3339 	if (exts->nr_actions == 0)
3340 		return NULL;
3341 	else
3342 		return exts->actions[0];
3343 }
3344 #endif
3345 
3346 int tcf_exts_dump(struct sk_buff *skb, struct tcf_exts *exts)
3347 {
3348 #ifdef CONFIG_NET_CLS_ACT
3349 	struct nlattr *nest;
3350 
3351 	if (exts->action && tcf_exts_has_actions(exts)) {
3352 		/*
3353 		 * again for backward compatible mode - we want
3354 		 * to work with both old and new modes of entering
3355 		 * tc data even if iproute2  was newer - jhs
3356 		 */
3357 		if (exts->type != TCA_OLD_COMPAT) {
3358 			nest = nla_nest_start_noflag(skb, exts->action);
3359 			if (nest == NULL)
3360 				goto nla_put_failure;
3361 
3362 			if (tcf_action_dump(skb, exts->actions, 0, 0, false)
3363 			    < 0)
3364 				goto nla_put_failure;
3365 			nla_nest_end(skb, nest);
3366 		} else if (exts->police) {
3367 			struct tc_action *act = tcf_exts_first_act(exts);
3368 			nest = nla_nest_start_noflag(skb, exts->police);
3369 			if (nest == NULL || !act)
3370 				goto nla_put_failure;
3371 			if (tcf_action_dump_old(skb, act, 0, 0) < 0)
3372 				goto nla_put_failure;
3373 			nla_nest_end(skb, nest);
3374 		}
3375 	}
3376 	return 0;
3377 
3378 nla_put_failure:
3379 	nla_nest_cancel(skb, nest);
3380 	return -1;
3381 #else
3382 	return 0;
3383 #endif
3384 }
3385 EXPORT_SYMBOL(tcf_exts_dump);
3386 
3387 int tcf_exts_terse_dump(struct sk_buff *skb, struct tcf_exts *exts)
3388 {
3389 #ifdef CONFIG_NET_CLS_ACT
3390 	struct nlattr *nest;
3391 
3392 	if (!exts->action || !tcf_exts_has_actions(exts))
3393 		return 0;
3394 
3395 	nest = nla_nest_start_noflag(skb, exts->action);
3396 	if (!nest)
3397 		goto nla_put_failure;
3398 
3399 	if (tcf_action_dump(skb, exts->actions, 0, 0, true) < 0)
3400 		goto nla_put_failure;
3401 	nla_nest_end(skb, nest);
3402 	return 0;
3403 
3404 nla_put_failure:
3405 	nla_nest_cancel(skb, nest);
3406 	return -1;
3407 #else
3408 	return 0;
3409 #endif
3410 }
3411 EXPORT_SYMBOL(tcf_exts_terse_dump);
3412 
3413 int tcf_exts_dump_stats(struct sk_buff *skb, struct tcf_exts *exts)
3414 {
3415 #ifdef CONFIG_NET_CLS_ACT
3416 	struct tc_action *a = tcf_exts_first_act(exts);
3417 	if (a != NULL && tcf_action_copy_stats(skb, a, 1) < 0)
3418 		return -1;
3419 #endif
3420 	return 0;
3421 }
3422 EXPORT_SYMBOL(tcf_exts_dump_stats);
3423 
3424 static void tcf_block_offload_inc(struct tcf_block *block, u32 *flags)
3425 {
3426 	if (*flags & TCA_CLS_FLAGS_IN_HW)
3427 		return;
3428 	*flags |= TCA_CLS_FLAGS_IN_HW;
3429 	atomic_inc(&block->offloadcnt);
3430 }
3431 
3432 static void tcf_block_offload_dec(struct tcf_block *block, u32 *flags)
3433 {
3434 	if (!(*flags & TCA_CLS_FLAGS_IN_HW))
3435 		return;
3436 	*flags &= ~TCA_CLS_FLAGS_IN_HW;
3437 	atomic_dec(&block->offloadcnt);
3438 }
3439 
3440 static void tc_cls_offload_cnt_update(struct tcf_block *block,
3441 				      struct tcf_proto *tp, u32 *cnt,
3442 				      u32 *flags, u32 diff, bool add)
3443 {
3444 	lockdep_assert_held(&block->cb_lock);
3445 
3446 	spin_lock(&tp->lock);
3447 	if (add) {
3448 		if (!*cnt)
3449 			tcf_block_offload_inc(block, flags);
3450 		*cnt += diff;
3451 	} else {
3452 		*cnt -= diff;
3453 		if (!*cnt)
3454 			tcf_block_offload_dec(block, flags);
3455 	}
3456 	spin_unlock(&tp->lock);
3457 }
3458 
3459 static void
3460 tc_cls_offload_cnt_reset(struct tcf_block *block, struct tcf_proto *tp,
3461 			 u32 *cnt, u32 *flags)
3462 {
3463 	lockdep_assert_held(&block->cb_lock);
3464 
3465 	spin_lock(&tp->lock);
3466 	tcf_block_offload_dec(block, flags);
3467 	*cnt = 0;
3468 	spin_unlock(&tp->lock);
3469 }
3470 
3471 static int
3472 __tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3473 		   void *type_data, bool err_stop)
3474 {
3475 	struct flow_block_cb *block_cb;
3476 	int ok_count = 0;
3477 	int err;
3478 
3479 	list_for_each_entry(block_cb, &block->flow_block.cb_list, list) {
3480 		err = block_cb->cb(type, type_data, block_cb->cb_priv);
3481 		if (err) {
3482 			if (err_stop)
3483 				return err;
3484 		} else {
3485 			ok_count++;
3486 		}
3487 	}
3488 	return ok_count;
3489 }
3490 
3491 int tc_setup_cb_call(struct tcf_block *block, enum tc_setup_type type,
3492 		     void *type_data, bool err_stop, bool rtnl_held)
3493 {
3494 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3495 	int ok_count;
3496 
3497 retry:
3498 	if (take_rtnl)
3499 		rtnl_lock();
3500 	down_read(&block->cb_lock);
3501 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3502 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3503 	 * obtain the locks in same order here.
3504 	 */
3505 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3506 		up_read(&block->cb_lock);
3507 		take_rtnl = true;
3508 		goto retry;
3509 	}
3510 
3511 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3512 
3513 	up_read(&block->cb_lock);
3514 	if (take_rtnl)
3515 		rtnl_unlock();
3516 	return ok_count;
3517 }
3518 EXPORT_SYMBOL(tc_setup_cb_call);
3519 
3520 /* Non-destructive filter add. If filter that wasn't already in hardware is
3521  * successfully offloaded, increment block offloads counter. On failure,
3522  * previously offloaded filter is considered to be intact and offloads counter
3523  * is not decremented.
3524  */
3525 
3526 int tc_setup_cb_add(struct tcf_block *block, struct tcf_proto *tp,
3527 		    enum tc_setup_type type, void *type_data, bool err_stop,
3528 		    u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3529 {
3530 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3531 	int ok_count;
3532 
3533 retry:
3534 	if (take_rtnl)
3535 		rtnl_lock();
3536 	down_read(&block->cb_lock);
3537 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3538 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3539 	 * obtain the locks in same order here.
3540 	 */
3541 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3542 		up_read(&block->cb_lock);
3543 		take_rtnl = true;
3544 		goto retry;
3545 	}
3546 
3547 	/* Make sure all netdevs sharing this block are offload-capable. */
3548 	if (block->nooffloaddevcnt && err_stop) {
3549 		ok_count = -EOPNOTSUPP;
3550 		goto err_unlock;
3551 	}
3552 
3553 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3554 	if (ok_count < 0)
3555 		goto err_unlock;
3556 
3557 	if (tp->ops->hw_add)
3558 		tp->ops->hw_add(tp, type_data);
3559 	if (ok_count > 0)
3560 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags,
3561 					  ok_count, true);
3562 err_unlock:
3563 	up_read(&block->cb_lock);
3564 	if (take_rtnl)
3565 		rtnl_unlock();
3566 	return min(ok_count, 0);
3567 }
3568 EXPORT_SYMBOL(tc_setup_cb_add);
3569 
3570 /* Destructive filter replace. If filter that wasn't already in hardware is
3571  * successfully offloaded, increment block offload counter. On failure,
3572  * previously offloaded filter is considered to be destroyed and offload counter
3573  * is decremented.
3574  */
3575 
3576 int tc_setup_cb_replace(struct tcf_block *block, struct tcf_proto *tp,
3577 			enum tc_setup_type type, void *type_data, bool err_stop,
3578 			u32 *old_flags, unsigned int *old_in_hw_count,
3579 			u32 *new_flags, unsigned int *new_in_hw_count,
3580 			bool rtnl_held)
3581 {
3582 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3583 	int ok_count;
3584 
3585 retry:
3586 	if (take_rtnl)
3587 		rtnl_lock();
3588 	down_read(&block->cb_lock);
3589 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3590 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3591 	 * obtain the locks in same order here.
3592 	 */
3593 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3594 		up_read(&block->cb_lock);
3595 		take_rtnl = true;
3596 		goto retry;
3597 	}
3598 
3599 	/* Make sure all netdevs sharing this block are offload-capable. */
3600 	if (block->nooffloaddevcnt && err_stop) {
3601 		ok_count = -EOPNOTSUPP;
3602 		goto err_unlock;
3603 	}
3604 
3605 	tc_cls_offload_cnt_reset(block, tp, old_in_hw_count, old_flags);
3606 	if (tp->ops->hw_del)
3607 		tp->ops->hw_del(tp, type_data);
3608 
3609 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3610 	if (ok_count < 0)
3611 		goto err_unlock;
3612 
3613 	if (tp->ops->hw_add)
3614 		tp->ops->hw_add(tp, type_data);
3615 	if (ok_count > 0)
3616 		tc_cls_offload_cnt_update(block, tp, new_in_hw_count,
3617 					  new_flags, ok_count, true);
3618 err_unlock:
3619 	up_read(&block->cb_lock);
3620 	if (take_rtnl)
3621 		rtnl_unlock();
3622 	return min(ok_count, 0);
3623 }
3624 EXPORT_SYMBOL(tc_setup_cb_replace);
3625 
3626 /* Destroy filter and decrement block offload counter, if filter was previously
3627  * offloaded.
3628  */
3629 
3630 int tc_setup_cb_destroy(struct tcf_block *block, struct tcf_proto *tp,
3631 			enum tc_setup_type type, void *type_data, bool err_stop,
3632 			u32 *flags, unsigned int *in_hw_count, bool rtnl_held)
3633 {
3634 	bool take_rtnl = READ_ONCE(block->lockeddevcnt) && !rtnl_held;
3635 	int ok_count;
3636 
3637 retry:
3638 	if (take_rtnl)
3639 		rtnl_lock();
3640 	down_read(&block->cb_lock);
3641 	/* Need to obtain rtnl lock if block is bound to devs that require it.
3642 	 * In block bind code cb_lock is obtained while holding rtnl, so we must
3643 	 * obtain the locks in same order here.
3644 	 */
3645 	if (!rtnl_held && !take_rtnl && block->lockeddevcnt) {
3646 		up_read(&block->cb_lock);
3647 		take_rtnl = true;
3648 		goto retry;
3649 	}
3650 
3651 	ok_count = __tc_setup_cb_call(block, type, type_data, err_stop);
3652 
3653 	tc_cls_offload_cnt_reset(block, tp, in_hw_count, flags);
3654 	if (tp->ops->hw_del)
3655 		tp->ops->hw_del(tp, type_data);
3656 
3657 	up_read(&block->cb_lock);
3658 	if (take_rtnl)
3659 		rtnl_unlock();
3660 	return min(ok_count, 0);
3661 }
3662 EXPORT_SYMBOL(tc_setup_cb_destroy);
3663 
3664 int tc_setup_cb_reoffload(struct tcf_block *block, struct tcf_proto *tp,
3665 			  bool add, flow_setup_cb_t *cb,
3666 			  enum tc_setup_type type, void *type_data,
3667 			  void *cb_priv, u32 *flags, unsigned int *in_hw_count)
3668 {
3669 	int err = cb(type, type_data, cb_priv);
3670 
3671 	if (err) {
3672 		if (add && tc_skip_sw(*flags))
3673 			return err;
3674 	} else {
3675 		tc_cls_offload_cnt_update(block, tp, in_hw_count, flags, 1,
3676 					  add);
3677 	}
3678 
3679 	return 0;
3680 }
3681 EXPORT_SYMBOL(tc_setup_cb_reoffload);
3682 
3683 static int tcf_act_get_user_cookie(struct flow_action_entry *entry,
3684 				   const struct tc_action *act)
3685 {
3686 	struct tc_cookie *user_cookie;
3687 	int err = 0;
3688 
3689 	rcu_read_lock();
3690 	user_cookie = rcu_dereference(act->user_cookie);
3691 	if (user_cookie) {
3692 		entry->user_cookie = flow_action_cookie_create(user_cookie->data,
3693 							       user_cookie->len,
3694 							       GFP_ATOMIC);
3695 		if (!entry->user_cookie)
3696 			err = -ENOMEM;
3697 	}
3698 	rcu_read_unlock();
3699 	return err;
3700 }
3701 
3702 static void tcf_act_put_user_cookie(struct flow_action_entry *entry)
3703 {
3704 	flow_action_cookie_destroy(entry->user_cookie);
3705 }
3706 
3707 void tc_cleanup_offload_action(struct flow_action *flow_action)
3708 {
3709 	struct flow_action_entry *entry;
3710 	int i;
3711 
3712 	flow_action_for_each(i, entry, flow_action) {
3713 		tcf_act_put_user_cookie(entry);
3714 		if (entry->destructor)
3715 			entry->destructor(entry->destructor_priv);
3716 	}
3717 }
3718 EXPORT_SYMBOL(tc_cleanup_offload_action);
3719 
3720 static int tc_setup_offload_act(struct tc_action *act,
3721 				struct flow_action_entry *entry,
3722 				u32 *index_inc,
3723 				struct netlink_ext_ack *extack)
3724 {
3725 #ifdef CONFIG_NET_CLS_ACT
3726 	if (act->ops->offload_act_setup) {
3727 		return act->ops->offload_act_setup(act, entry, index_inc, true,
3728 						   extack);
3729 	} else {
3730 		NL_SET_ERR_MSG(extack, "Action does not support offload");
3731 		return -EOPNOTSUPP;
3732 	}
3733 #else
3734 	return 0;
3735 #endif
3736 }
3737 
3738 int tc_setup_action(struct flow_action *flow_action,
3739 		    struct tc_action *actions[],
3740 		    u32 miss_cookie_base,
3741 		    struct netlink_ext_ack *extack)
3742 {
3743 	int i, j, k, index, err = 0;
3744 	struct tc_action *act;
3745 
3746 	BUILD_BUG_ON(TCA_ACT_HW_STATS_ANY != FLOW_ACTION_HW_STATS_ANY);
3747 	BUILD_BUG_ON(TCA_ACT_HW_STATS_IMMEDIATE != FLOW_ACTION_HW_STATS_IMMEDIATE);
3748 	BUILD_BUG_ON(TCA_ACT_HW_STATS_DELAYED != FLOW_ACTION_HW_STATS_DELAYED);
3749 
3750 	if (!actions)
3751 		return 0;
3752 
3753 	j = 0;
3754 	tcf_act_for_each_action(i, act, actions) {
3755 		struct flow_action_entry *entry;
3756 
3757 		entry = &flow_action->entries[j];
3758 		spin_lock_bh(&act->tcfa_lock);
3759 		err = tcf_act_get_user_cookie(entry, act);
3760 		if (err)
3761 			goto err_out_locked;
3762 
3763 		index = 0;
3764 		err = tc_setup_offload_act(act, entry, &index, extack);
3765 		if (err)
3766 			goto err_out_locked;
3767 
3768 		for (k = 0; k < index ; k++) {
3769 			entry[k].hw_stats = tc_act_hw_stats(act->hw_stats);
3770 			entry[k].hw_index = act->tcfa_index;
3771 			entry[k].cookie = (unsigned long)act;
3772 			entry[k].miss_cookie =
3773 				tcf_exts_miss_cookie_get(miss_cookie_base, i);
3774 		}
3775 
3776 		j += index;
3777 
3778 		spin_unlock_bh(&act->tcfa_lock);
3779 	}
3780 
3781 err_out:
3782 	if (err)
3783 		tc_cleanup_offload_action(flow_action);
3784 
3785 	return err;
3786 err_out_locked:
3787 	spin_unlock_bh(&act->tcfa_lock);
3788 	goto err_out;
3789 }
3790 
3791 int tc_setup_offload_action(struct flow_action *flow_action,
3792 			    const struct tcf_exts *exts,
3793 			    struct netlink_ext_ack *extack)
3794 {
3795 #ifdef CONFIG_NET_CLS_ACT
3796 	u32 miss_cookie_base;
3797 
3798 	if (!exts)
3799 		return 0;
3800 
3801 	miss_cookie_base = exts->miss_cookie_node ?
3802 			   exts->miss_cookie_node->miss_cookie_base : 0;
3803 	return tc_setup_action(flow_action, exts->actions, miss_cookie_base,
3804 			       extack);
3805 #else
3806 	return 0;
3807 #endif
3808 }
3809 EXPORT_SYMBOL(tc_setup_offload_action);
3810 
3811 unsigned int tcf_exts_num_actions(struct tcf_exts *exts)
3812 {
3813 	unsigned int num_acts = 0;
3814 	struct tc_action *act;
3815 	int i;
3816 
3817 	tcf_exts_for_each_action(i, act, exts) {
3818 		if (is_tcf_pedit(act))
3819 			num_acts += tcf_pedit_nkeys(act);
3820 		else
3821 			num_acts++;
3822 	}
3823 	return num_acts;
3824 }
3825 EXPORT_SYMBOL(tcf_exts_num_actions);
3826 
3827 #ifdef CONFIG_NET_CLS_ACT
3828 static int tcf_qevent_parse_block_index(struct nlattr *block_index_attr,
3829 					u32 *p_block_index,
3830 					struct netlink_ext_ack *extack)
3831 {
3832 	*p_block_index = nla_get_u32(block_index_attr);
3833 	if (!*p_block_index) {
3834 		NL_SET_ERR_MSG(extack, "Block number may not be zero");
3835 		return -EINVAL;
3836 	}
3837 
3838 	return 0;
3839 }
3840 
3841 int tcf_qevent_init(struct tcf_qevent *qe, struct Qdisc *sch,
3842 		    enum flow_block_binder_type binder_type,
3843 		    struct nlattr *block_index_attr,
3844 		    struct netlink_ext_ack *extack)
3845 {
3846 	u32 block_index;
3847 	int err;
3848 
3849 	if (!block_index_attr)
3850 		return 0;
3851 
3852 	err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3853 	if (err)
3854 		return err;
3855 
3856 	qe->info.binder_type = binder_type;
3857 	qe->info.chain_head_change = tcf_chain_head_change_dflt;
3858 	qe->info.chain_head_change_priv = &qe->filter_chain;
3859 	qe->info.block_index = block_index;
3860 
3861 	return tcf_block_get_ext(&qe->block, sch, &qe->info, extack);
3862 }
3863 EXPORT_SYMBOL(tcf_qevent_init);
3864 
3865 void tcf_qevent_destroy(struct tcf_qevent *qe, struct Qdisc *sch)
3866 {
3867 	if (qe->info.block_index)
3868 		tcf_block_put_ext(qe->block, sch, &qe->info);
3869 }
3870 EXPORT_SYMBOL(tcf_qevent_destroy);
3871 
3872 int tcf_qevent_validate_change(struct tcf_qevent *qe, struct nlattr *block_index_attr,
3873 			       struct netlink_ext_ack *extack)
3874 {
3875 	u32 block_index;
3876 	int err;
3877 
3878 	if (!block_index_attr)
3879 		return 0;
3880 
3881 	err = tcf_qevent_parse_block_index(block_index_attr, &block_index, extack);
3882 	if (err)
3883 		return err;
3884 
3885 	/* Bounce newly-configured block or change in block. */
3886 	if (block_index != qe->info.block_index) {
3887 		NL_SET_ERR_MSG(extack, "Change of blocks is not supported");
3888 		return -EINVAL;
3889 	}
3890 
3891 	return 0;
3892 }
3893 EXPORT_SYMBOL(tcf_qevent_validate_change);
3894 
3895 struct sk_buff *tcf_qevent_handle(struct tcf_qevent *qe, struct Qdisc *sch, struct sk_buff *skb,
3896 				  struct sk_buff **to_free, int *ret)
3897 {
3898 	struct tcf_result cl_res;
3899 	struct tcf_proto *fl;
3900 
3901 	if (!qe->info.block_index)
3902 		return skb;
3903 
3904 	fl = rcu_dereference_bh(qe->filter_chain);
3905 
3906 	switch (tcf_classify(skb, NULL, fl, &cl_res, false)) {
3907 	case TC_ACT_SHOT:
3908 		qdisc_qstats_drop(sch);
3909 		__qdisc_drop(skb, to_free);
3910 		*ret = __NET_XMIT_BYPASS;
3911 		return NULL;
3912 	case TC_ACT_STOLEN:
3913 	case TC_ACT_QUEUED:
3914 	case TC_ACT_TRAP:
3915 		__qdisc_drop(skb, to_free);
3916 		*ret = __NET_XMIT_STOLEN;
3917 		return NULL;
3918 	case TC_ACT_REDIRECT:
3919 		skb_do_redirect(skb);
3920 		*ret = __NET_XMIT_STOLEN;
3921 		return NULL;
3922 	}
3923 
3924 	return skb;
3925 }
3926 EXPORT_SYMBOL(tcf_qevent_handle);
3927 
3928 int tcf_qevent_dump(struct sk_buff *skb, int attr_name, struct tcf_qevent *qe)
3929 {
3930 	if (!qe->info.block_index)
3931 		return 0;
3932 	return nla_put_u32(skb, attr_name, qe->info.block_index);
3933 }
3934 EXPORT_SYMBOL(tcf_qevent_dump);
3935 #endif
3936 
3937 static __net_init int tcf_net_init(struct net *net)
3938 {
3939 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3940 
3941 	spin_lock_init(&tn->idr_lock);
3942 	idr_init(&tn->idr);
3943 	return 0;
3944 }
3945 
3946 static void __net_exit tcf_net_exit(struct net *net)
3947 {
3948 	struct tcf_net *tn = net_generic(net, tcf_net_id);
3949 
3950 	idr_destroy(&tn->idr);
3951 }
3952 
3953 static struct pernet_operations tcf_net_ops = {
3954 	.init = tcf_net_init,
3955 	.exit = tcf_net_exit,
3956 	.id   = &tcf_net_id,
3957 	.size = sizeof(struct tcf_net),
3958 };
3959 
3960 static int __init tc_filter_init(void)
3961 {
3962 	int err;
3963 
3964 	tc_filter_wq = alloc_ordered_workqueue("tc_filter_workqueue", 0);
3965 	if (!tc_filter_wq)
3966 		return -ENOMEM;
3967 
3968 	err = register_pernet_subsys(&tcf_net_ops);
3969 	if (err)
3970 		goto err_register_pernet_subsys;
3971 
3972 	xa_init_flags(&tcf_exts_miss_cookies_xa, XA_FLAGS_ALLOC1);
3973 
3974 	rtnl_register(PF_UNSPEC, RTM_NEWTFILTER, tc_new_tfilter, NULL,
3975 		      RTNL_FLAG_DOIT_UNLOCKED);
3976 	rtnl_register(PF_UNSPEC, RTM_DELTFILTER, tc_del_tfilter, NULL,
3977 		      RTNL_FLAG_DOIT_UNLOCKED);
3978 	rtnl_register(PF_UNSPEC, RTM_GETTFILTER, tc_get_tfilter,
3979 		      tc_dump_tfilter, RTNL_FLAG_DOIT_UNLOCKED);
3980 	rtnl_register(PF_UNSPEC, RTM_NEWCHAIN, tc_ctl_chain, NULL, 0);
3981 	rtnl_register(PF_UNSPEC, RTM_DELCHAIN, tc_ctl_chain, NULL, 0);
3982 	rtnl_register(PF_UNSPEC, RTM_GETCHAIN, tc_ctl_chain,
3983 		      tc_dump_chain, 0);
3984 
3985 	return 0;
3986 
3987 err_register_pernet_subsys:
3988 	destroy_workqueue(tc_filter_wq);
3989 	return err;
3990 }
3991 
3992 subsys_initcall(tc_filter_init);
3993